@houtini/lm 2.1.0 → 2.4.1

package/README.md

@@ -3,39 +3,50 @@
 [![npm version](https://img.shields.io/npm/v/@houtini/lm)](https://www.npmjs.com/package/@houtini/lm)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 
-MCP server that connects Claude to **any OpenAI-compatible LLM endpoint** — LM Studio, Ollama, vLLM, llama.cpp, or any remote API.
+I built this because I kept leaving Claude Code running overnight on big refactors and the token bill was painful. A huge chunk of that spend goes on bounded tasks any decent model handles fine - generating boilerplate, explaining code, drafting commit messages, converting formats. Stuff that doesn't need Claude's reasoning or tool access.
 
-Offload routine work to a local model. Keep your Claude context window for the hard stuff.
+Houtini LM connects Claude Code to a local LLM on your network. Claude keeps doing the hard work - architecture, planning, multi-file changes - and offloads the grunt work to your local model. Free. No rate limits. Private.
 
-## Why
+The session footer tracks everything Claude offloads, so you can watch the savings stack up.
 
-Claude is great at orchestration and reasoning. Local models are great at bulk analysis, classification, extraction, and summarisation. This server lets Claude delegate to a local model on the fly — no API keys, no cloud round-trips, no context wasted.
+## How it works
 
-**Common use cases:**
+```
+Claude Code (orchestrator)
+   │
+   ├─ Complex reasoning, planning, architecture → Claude API (your tokens)
+   │
+   └─ Bounded grunt work → houtini-lm ──HTTP/SSE──> Your local LLM (free)
+       • Boilerplate & test stubs          Qwen, Llama, Mistral, DeepSeek...
+       • Code review & explanations        LM Studio, Ollama, vLLM, llama.cpp
+       • Commit messages & docs
+       • Format conversion
+       • Mock data & type definitions
+```
+
+Claude's the architect. Your local model's the drafter. Claude QAs everything.
 
-- Classify or tag hundreds of items without burning Claude tokens
-- Extract structured data from long documents
-- Run a second opinion on generated code
-- Summarise research before Claude synthesises it
-- Delegate code review to a local model while Claude handles other work
+## Quick start
 
-## What's new in v2.1.0
+### Claude Code
+
+```bash
+claude mcp add houtini-lm -- npx -y @houtini/lm
+```
 
-- **Smarter tool descriptions** — tool descriptions now encode prompting best practices for local LLMs, so Claude automatically sends well-structured prompts (complete code, capped output tokens, explicit format instructions)
-- **New `code_task` tool** — purpose-built for code analysis with an optimised system prompt and sensible defaults (temp 0.2, 500 token cap)
-- **Delegation guidance** — each tool description tells Claude when to use it, what output to expect, and what to avoid (e.g. never send truncated code to a local model)
+That's it. If LM Studio's running on `localhost:1234` (the default), Claude can start delegating straight away.
 
-## Install
+### LLM on a different machine
 
-### Claude Code (recommended)
+I've got a GPU box on my local network running Qwen 3 Coder Next in LM Studio. If you've got a similar setup, point the URL at it:
 
 ```bash
-claude mcp add houtini-lm -e LM_STUDIO_URL=http://localhost:1234 -- npx -y @houtini/lm
+claude mcp add houtini-lm -e LM_STUDIO_URL=http://192.168.1.50:1234 -- npx -y @houtini/lm
 ```
 
 ### Claude Desktop
 
-Add to `claude_desktop_config.json`:
+Drop this into your `claude_desktop_config.json`:
 
 ```json
 {
@@ -51,93 +62,127 @@ Add to `claude_desktop_config.json`:
 }
 ```
 
-### npx (standalone)
+## What gets offloaded
 
-```bash
-npx @houtini/lm
-```
+**Delegate to the local model** - bounded, well-defined tasks:
 
-## Configuration
+| Task | Why it works locally |
+|------|---------------------|
+| Generate test stubs | Clear input (source), clear output (tests) |
+| Explain a function | Summarisation doesn't need tool access |
+| Draft commit messages | Diff in, message out |
+| Code review | Paste full source, ask for bugs |
+| Convert formats | JSON↔YAML, snake_case↔camelCase |
+| Generate mock data | Schema in, data out |
+| Write type definitions | Source in, types out |
+| Brainstorm approaches | Doesn't commit to anything |
 
-Set via environment variables or in your MCP client config:
+**Keep on Claude** - anything that needs reasoning, tool access, or multi-step orchestration:
 
-| Variable | Default | Description |
-|----------|---------|-------------|
-| `LM_STUDIO_URL` | `http://localhost:1234` | Base URL of the OpenAI-compatible API |
-| `LM_STUDIO_MODEL` | *(auto-detect)* | Model identifier — leave blank to use whatever's loaded |
-| `LM_STUDIO_PASSWORD` | *(none)* | Bearer token for authenticated endpoints |
+- Architectural decisions
+- Reading/writing files
+- Running tests and interpreting results
+- Multi-file refactoring plans
+- Anything that needs to call other tools
 
-## Tools
+The tool descriptions are written to nudge Claude into planning delegation at the start of large tasks, not just using it when it happens to think of it.
 
-### `chat`
+## Token tracking
 
-Delegate a bounded task to the local LLM. The workhorse for quick questions, code explanation, and pattern recognition.
+Every response includes a session footer:
 
 ```
-message (required) — the task, with explicit output format instructions
-system             — persona (be specific: "Senior TypeScript dev", not "helpful assistant")
-temperature        — 0.1 for code, 0.3 for analysis (default), 0.5 for suggestions
-max_tokens         — match to expected output: 150 for quick answers, 300 for explanations, 500 for code gen
+Model: qwen/qwen3-coder-next | This call: 145→248 tokens | Session: 12,450 tokens offloaded across 23 calls
 ```
 
-**Tip:** Always send complete code — local models hallucinate details for truncated input.
+The `discover` tool reports cumulative session stats too. Claude sees this data and (I've found) it reinforces the delegation habit throughout long-running tasks. The more it sees it's saving tokens, the more it looks for things to offload.
+
+## Tools
+
+### `chat`
+
+The workhorse. Send a task, get an answer. The description includes planning triggers that nudge Claude to identify offloadable work when it's starting a big task.
+
+| Parameter | Required | Default | What it does |
+|-----------|----------|---------|-------------|
+| `message` | yes | - | The task. Be specific about output format. |
+| `system` | no | - | Persona - "Senior TypeScript dev" not "helpful assistant" |
+| `temperature` | no | 0.3 | 0.1 for code, 0.3 for analysis, 0.7 for creative |
+| `max_tokens` | no | 2048 | Lower for quick answers, higher for generation |
 
 ### `custom_prompt`
 
-Structured 3-part prompt with separate system, context, and instruction fields. The separation prevents context bleed in local models — better results than stuffing everything into a single message.
+Three-part prompt: system, context, instruction. Keeping them separate prevents context bleed - consistently outperforms stuffing everything into one message, especially with local models.
 
-```
-instruction (required) — what to produce (under 50 words works best)
-system                 — persona, specific and under 30 words
-context                — COMPLETE data to analyse (never truncated)
-temperature            — 0.1 for review, 0.3 for analysis (default)
-max_tokens             — 200 for bullets, 400 for detailed review, 600 for code gen
-```
+| Parameter | Required | Default | What it does |
+|-----------|----------|---------|-------------|
+| `instruction` | yes | - | What to produce. Under 50 words works best. |
+| `system` | no | - | Persona + constraints, under 30 words |
+| `context` | no | - | Complete data to analyse. Never truncate. |
+| `temperature` | no | 0.3 | 0.1 for review, 0.3 for analysis |
+| `max_tokens` | no | 2048 | Match to expected output length |
 
 ### `code_task`
 
-Purpose-built for code analysis. Wraps the local LLM with an optimised code-review system prompt and low temperature (0.2).
+Built for code analysis. Pre-configured system prompt, locked to temperature 0.2 for focused output.
 
-```
-code (required)     — complete source code (never truncate)
-task (required)     — what to do: "Find bugs", "Explain this function", "Add error handling"
-language            — "typescript", "python", "rust", etc.
-max_tokens          — default 500 (200 for quick answers, 800 for code generation)
-```
+| Parameter | Required | Default | What it does |
+|-----------|----------|---------|-------------|
+| `code` | yes | - | Complete source code. Never truncate. |
+| `task` | yes | - | "Find bugs", "Explain this", "Write tests" |
+| `language` | no | - | "typescript", "python", "rust", etc. |
+| `max_tokens` | no | 2048 | Match to expected output length |
 
-**The local LLM excels at:** explaining code, finding common bugs, suggesting improvements, comparing patterns, generating boilerplate.
+### `discover`
 
-**It struggles with:** subtle/adversarial bugs, multi-file reasoning, design tasks requiring integration.
+Health check. Returns model name, context window, latency, and cumulative session stats. Call before delegating if you're not sure the LLM's available.
 
 ### `list_models`
 
-Returns the models currently loaded on the LLM server.
+Lists everything loaded on the LLM server with context window sizes.
+
+## Getting good results from local models
 
-### `health_check`
+Qwen, Llama, DeepSeek - they score brilliantly on coding benchmarks now. The gap between a good and bad result is almost always **prompt quality**, not model capability. I've spent a fair bit of time on this.
 
-Checks connectivity. Returns response time, auth status, and loaded model count.
+**Send complete code.** Local models hallucinate details when you give them truncated input. If a file's too large, send the relevant function - not a snippet with `...` in the middle.
 
-## Performance guide
+**Be explicit about output format.** "Return a JSON array" or "respond in bullet points" - don't leave it open-ended. Smaller models need this.
 
-At typical local LLM speeds (~3-4 tokens/second on consumer hardware):
+**Set a specific persona.** "Expert Rust developer who cares about memory safety" gets noticeably better results than "helpful assistant."
 
-| max_tokens | Response time | Best for |
-|------------|--------------|----------|
-| 150 | ~45 seconds | Quick questions, classifications |
-| 300 | ~100 seconds | Code explanations, summaries |
-| 500 | ~170 seconds | Code review, generation |
+**State constraints.** "No preamble", "reference line numbers", "max 5 bullet points" - tell the model what *not* to do as well as what to do.
 
-Set `max_tokens` to match your expected output — lower values mean faster responses.
+**Include surrounding context.** For code generation, send imports, types, and function signatures - not just the function body.
+
+**One call at a time.** If your LLM server runs a single model, parallel calls queue up and stack timeouts. Send them sequentially.
+
+## Configuration
+
+| Variable | Default | What it does |
+|----------|---------|-------------|
+| `LM_STUDIO_URL` | `http://localhost:1234` | Base URL of the OpenAI-compatible API |
+| `LM_STUDIO_MODEL` | *(auto-detect)* | Model identifier - leave blank to use whatever's loaded |
+| `LM_STUDIO_PASSWORD` | *(none)* | Bearer token for authenticated endpoints |
+| `LM_CONTEXT_WINDOW` | `100000` | Fallback context window if the API doesn't report it |
 
 ## Compatible endpoints
 
-| Provider | URL | Notes |
-|----------|-----|-------|
+Works with anything that speaks the OpenAI `/v1/chat/completions` API:
+
+| What | URL | Notes |
+|------|-----|-------|
 | [LM Studio](https://lmstudio.ai) | `http://localhost:1234` | Default, zero config |
-| [Ollama](https://ollama.com) | `http://localhost:11434` | Use OpenAI-compatible mode |
+| [Ollama](https://ollama.com) | `http://localhost:11434` | Set `LM_STUDIO_URL` |
 | [vLLM](https://docs.vllm.ai) | `http://localhost:8000` | Native OpenAI API |
 | [llama.cpp](https://github.com/ggml-org/llama.cpp) | `http://localhost:8080` | Server mode |
-| Remote / cloud APIs | Any URL | Set `LM_STUDIO_URL` + `LM_STUDIO_PASSWORD` |
+| Any OpenAI-compatible API | Any URL | Set URL + password |
+
+## Streaming and timeouts
+
+All inference uses Server-Sent Events streaming. Tokens arrive incrementally, keeping the connection alive. If generation takes longer than 55 seconds, you get a partial result instead of a timeout error - the footer shows `⚠ TRUNCATED` when this happens.
+
+The 55-second soft timeout exists because the MCP SDK has a hard ~60s client-side timeout. Without streaming, any response that took longer than 60 seconds just vanished. Not ideal.
 
 ## Development
 
@@ -148,12 +193,6 @@ npm install
 npm run build
 ```
 
-Run the test suite against a live LLM server:
-
-```bash
-node test.mjs
-```
-
-## License
+## Licence
 
 MIT

package/dist/index.d.ts

@@ -3,6 +3,6 @@
  * Houtini LM — MCP Server for Local LLMs via OpenAI-compatible API
  *
  * Connects to LM Studio (or any OpenAI-compatible endpoint) and exposes
- * chat, custom prompts, and model info as MCP tools.
+ * chat, custom prompts, code tasks, and model discovery as MCP tools.
  */
 export {};

package/dist/index.js

@@ -3,7 +3,7 @@
  * Houtini LM — MCP Server for Local LLMs via OpenAI-compatible API
  *
  * Connects to LM Studio (or any OpenAI-compatible endpoint) and exposes
- * chat, custom prompts, and model info as MCP tools.
+ * chat, custom prompts, code tasks, and model discovery as MCP tools.
  */
 import { Server } from '@modelcontextprotocol/sdk/server/index.js';
 import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
@@ -11,117 +11,343 @@ import { CallToolRequestSchema, ListToolsRequestSchema, } from '@modelcontextpro
 const LM_BASE_URL = process.env.LM_STUDIO_URL || 'http://localhost:1234';
 const LM_MODEL = process.env.LM_STUDIO_MODEL || '';
 const LM_PASSWORD = process.env.LM_STUDIO_PASSWORD || '';
-const DEFAULT_MAX_TOKENS = 4096;
+const DEFAULT_MAX_TOKENS = 2048;
 const DEFAULT_TEMPERATURE = 0.3;
+const CONNECT_TIMEOUT_MS = 5000;
+const INFERENCE_CONNECT_TIMEOUT_MS = 30_000; // generous connect timeout for inference
+const SOFT_TIMEOUT_MS = 55_000; // return partial results before MCP SDK ~60s timeout
+const READ_CHUNK_TIMEOUT_MS = 30_000; // max wait for a single SSE chunk
+const FALLBACK_CONTEXT_LENGTH = parseInt(process.env.LM_CONTEXT_WINDOW || '100000', 10);
+// ── Session-level token accounting ───────────────────────────────────
+// Tracks cumulative tokens offloaded to the local LLM across all calls
+// in this session. Shown in every response footer so Claude can reason
+// about cost savings and continue delegating strategically.
+const session = {
+    calls: 0,
+    promptTokens: 0,
+    completionTokens: 0,
+};
+function recordUsage(usage) {
+    session.calls++;
+    if (usage) {
+        session.promptTokens += usage.prompt_tokens;
+        session.completionTokens += usage.completion_tokens;
+    }
+}
+function sessionSummary() {
+    const total = session.promptTokens + session.completionTokens;
+    if (session.calls === 0)
+        return '';
+    return `Session: ${total.toLocaleString()} tokens offloaded across ${session.calls} call${session.calls === 1 ? '' : 's'}`;
+}
 function apiHeaders() {
     const h = { 'Content-Type': 'application/json' };
     if (LM_PASSWORD)
         h['Authorization'] = `Bearer ${LM_PASSWORD}`;
     return h;
 }
-async function chatCompletion(messages, options = {}) {
+/**
+ * Fetch with a connect timeout so Claude doesn't hang when the host is offline.
+ */
+async function fetchWithTimeout(url, options, timeoutMs = CONNECT_TIMEOUT_MS) {
+    const controller = new AbortController();
+    const timer = setTimeout(() => controller.abort(), timeoutMs);
+    try {
+        return await fetch(url, { ...options, signal: controller.signal });
+    }
+    finally {
+        clearTimeout(timer);
+    }
+}
+/**
+ * Read from a stream with a per-chunk timeout.
+ * Prevents hanging forever if the LLM stalls mid-generation.
+ */
+async function timedRead(reader, timeoutMs) {
+    let timer;
+    const timeout = new Promise((resolve) => {
+        timer = setTimeout(() => resolve('timeout'), timeoutMs);
+    });
+    try {
+        return await Promise.race([reader.read(), timeout]);
+    }
+    finally {
+        clearTimeout(timer);
+    }
+}
+/**
+ * Streaming chat completion with soft timeout.
+ *
+ * Uses SSE streaming (`stream: true`) so tokens arrive incrementally.
+ * If we approach the MCP SDK's ~60s timeout (soft limit at 55s), we
+ * return whatever content we have so far with `truncated: true`.
+ * This means large code reviews return partial results instead of nothing.
+ */
+async function chatCompletionStreaming(messages, options = {}) {
     const body = {
         messages,
         temperature: options.temperature ?? DEFAULT_TEMPERATURE,
         max_tokens: options.maxTokens ?? DEFAULT_MAX_TOKENS,
-        stream: false,
+        stream: true,
     };
     if (options.model || LM_MODEL) {
         body.model = options.model || LM_MODEL;
     }
-    const res = await fetch(`${LM_BASE_URL}/v1/chat/completions`, {
-        method: 'POST',
-        headers: apiHeaders(),
-        body: JSON.stringify(body),
-    });
+    const startTime = Date.now();
+    const res = await fetchWithTimeout(`${LM_BASE_URL}/v1/chat/completions`, { method: 'POST', headers: apiHeaders(), body: JSON.stringify(body) }, INFERENCE_CONNECT_TIMEOUT_MS);
     if (!res.ok) {
         const text = await res.text().catch(() => '');
         throw new Error(`LM Studio API error ${res.status}: ${text}`);
     }
-    return res.json();
+    if (!res.body) {
+        throw new Error('Response body is null — streaming not supported by endpoint');
+    }
+    const reader = res.body.getReader();
+    const decoder = new TextDecoder();
+    let content = '';
+    let model = '';
+    let usage;
+    let finishReason = '';
+    let truncated = false;
+    let buffer = '';
+    try {
+        while (true) {
+            // Check soft timeout before each read
+            const elapsed = Date.now() - startTime;
+            if (elapsed > SOFT_TIMEOUT_MS) {
+                truncated = true;
+                process.stderr.write(`[houtini-lm] Soft timeout at ${elapsed}ms, returning ${content.length} chars of partial content\n`);
+                break;
+            }
+            // Read with per-chunk timeout (handles stalled generation)
+            const remaining = SOFT_TIMEOUT_MS - elapsed;
+            const chunkTimeout = Math.min(READ_CHUNK_TIMEOUT_MS, remaining);
+            const result = await timedRead(reader, chunkTimeout);
+            if (result === 'timeout') {
+                truncated = true;
+                process.stderr.write(`[houtini-lm] Chunk read timeout, returning ${content.length} chars of partial content\n`);
+                break;
+            }
+            if (result.done)
+                break;
+            buffer += decoder.decode(result.value, { stream: true });
+            // Parse SSE lines
+            const lines = buffer.split('\n');
+            buffer = lines.pop() || ''; // Keep incomplete line in buffer
+            for (const line of lines) {
+                const trimmed = line.trim();
+                if (!trimmed || trimmed === 'data: [DONE]')
+                    continue;
+                if (!trimmed.startsWith('data: '))
+                    continue;
+                try {
+                    const json = JSON.parse(trimmed.slice(6));
+                    if (json.model)
+                        model = json.model;
+                    const delta = json.choices?.[0]?.delta;
+                    if (delta?.content)
+                        content += delta.content;
+                    const reason = json.choices?.[0]?.finish_reason;
+                    if (reason)
+                        finishReason = reason;
+                    // Some endpoints include usage in the final streaming chunk
+                    if (json.usage)
+                        usage = json.usage;
+                }
+                catch {
+                    // Skip unparseable chunks (partial JSON, comments, etc.)
+                }
+            }
+        }
+    }
+    finally {
+        // Release the reader — don't await cancel() as it can hang
+        reader.releaseLock();
+    }
+    return { content, model, usage, finishReason, truncated };
 }
-async function listModels() {
-    const res = await fetch(`${LM_BASE_URL}/v1/models`, { headers: apiHeaders() });
+async function listModelsRaw() {
+    const res = await fetchWithTimeout(`${LM_BASE_URL}/v1/models`, { headers: apiHeaders() });
     if (!res.ok)
         throw new Error(`Failed to list models: ${res.status}`);
     const data = (await res.json());
-    return data.data.map((m) => m.id);
+    return data.data;
+}
+function getContextLength(model) {
+    // LM Studio uses context_length, vLLM uses max_model_len, fall back to env/100k
+    return model.context_length ?? model.max_model_len ?? FALLBACK_CONTEXT_LENGTH;
+}
+/**
+ * Format a footer line for streaming results showing model, usage, and truncation status.
+ */
+function formatFooter(resp, extra) {
+    // Record usage for session tracking before formatting
+    recordUsage(resp.usage);
+    const parts = [];
+    if (resp.model)
+        parts.push(`Model: ${resp.model}`);
+    if (resp.usage)
+        parts.push(`This call: ${resp.usage.prompt_tokens}→${resp.usage.completion_tokens} tokens`);
+    if (extra)
+        parts.push(extra);
+    if (resp.truncated)
+        parts.push('⚠ TRUNCATED (soft timeout — partial result)');
+    const sessionLine = sessionSummary();
+    if (sessionLine)
+        parts.push(sessionLine);
+    return parts.length > 0 ? `\n\n---\n${parts.join(' | ')}` : '';
 }
 // ── MCP Tool definitions ─────────────────────────────────────────────
 const TOOLS = [
     {
         name: 'chat',
-        description: 'Delegate a bounded task to the local LLM (Qwen3-Coder, ~3-4 tok/s). ' +
-            'Best for: quick code explanation, pattern recognition, boilerplate generation, knowledge questions. ' +
-            'Use when you can work in parallel — fire this off and continue your own work. ' +
-            'RULES: (1) Always send COMPLETE code, never truncated — the local LLM WILL hallucinate details for missing code. ' +
-            '(2) Set max_tokens to match expected output: 150 for quick answers (~45s), 300 for explanations (~100s), 500 for code generation (~170s). ' +
-            '(3) Be explicit about output format in your message ("respond in 3 bullets", "return only the function"). ' +
-            'DO NOT use for: multi-step reasoning, creative writing, tasks needing >500 token output, or anything requiring tool use (use lm-taskrunner for tool-augmented tasks via mcpo).',
+        description: 'Send a task to a local LLM running on a separate machine. This is a FREE, parallel worker — ' +
+            'use it to offload bounded work while you continue doing other things. The local LLM runs independently ' +
+            'and does not consume your tokens or rate limits.\n\n' +
+            'PLANNING: When you start a large task (refactoring, migrations, test suites, documentation), ' +
+            'break it into steps and identify which ones are bounded grunt work you can delegate here. ' +
+            'The more you offload, the more tokens you save. The session footer tracks cumulative savings.\n\n' +
+            'WHEN TO USE (delegate generously — it costs nothing):\n' +
+            '• Explain or summarise code/docs you just read\n' +
+            '• Generate boilerplate, test stubs, type definitions, mock data\n' +
+            '• Answer factual questions about languages, frameworks, APIs\n' +
+            '• Draft commit messages, PR descriptions, comments\n' +
+            '• Translate or reformat content (JSON↔YAML, snake_case↔camelCase)\n' +
+            '• Brainstorm approaches before you commit to one\n' +
+            '• Any self-contained subtask that does not need tool access\n\n' +
+            'PROMPT QUALITY (the local model is highly capable — results depend on your prompt):\n' +
+            '(1) Always send COMPLETE code/context — never truncate, the local LLM cannot access files.\n' +
+            '(2) Be explicit about output format ("respond as a JSON array", "return only the function").\n' +
+            '(3) Set a specific persona in the system field — "Senior TypeScript dev" beats "helpful assistant".\n' +
+            '(4) State constraints: "no preamble", "reference line numbers", "max 5 bullet points".\n' +
+            '(5) For code generation, include the surrounding context (imports, types, function signatures).\n\n' +
+            'QA: Always review the local LLM\'s output before using it. Verify correctness, check edge cases, ' +
+            'and fix any issues. You are the architect — the local model is a fast drafter, not the final authority.\n\n' +
+            'The local model, context window, and speed vary — call the discover tool to check what is loaded.',
         inputSchema: {
             type: 'object',
             properties: {
-                message: { type: 'string', description: 'The task. Be specific about expected output format and length. Include COMPLETE code — never truncate.' },
-                system: { type: 'string', description: 'Persona for the local LLM. Be specific: "Senior TypeScript dev" not "helpful assistant". Short personas (under 30 words) get best results.' },
-                temperature: { type: 'number', description: '0.1 for factual/code tasks, 0.3 for analysis (default), 0.7 for creative suggestions. Stay under 0.5 for code.' },
-                max_tokens: { type: 'number', description: 'Cap this to match expected output. 150=quick answer, 300=explanation, 500=code generation. Lower = faster. Default 4096 is almost always too high.' },
+                message: {
+                    type: 'string',
+                    description: 'The task. Be specific about expected output format. Include COMPLETE code/context — never truncate.',
+                },
+                system: {
+                    type: 'string',
+                    description: 'Persona for the local LLM. Be specific: "Senior TypeScript dev" not "helpful assistant".',
+                },
+                temperature: {
+                    type: 'number',
+                    description: '0.1 for factual/code, 0.3 for analysis (default), 0.7 for creative. Stay under 0.5 for code.',
+                },
+                max_tokens: {
+                    type: 'number',
+                    description: 'Max response tokens. Default 2048. Use higher for code generation, lower for quick answers.',
+                },
             },
             required: ['message'],
         },
     },
     {
         name: 'custom_prompt',
-        description: 'Structured analysis on the local LLM with explicit system/context/instruction separation. ' +
-            'This 3-part format gets the best results from local models — the separation prevents context bleed. ' +
-            'System sets persona (be specific: "Senior TypeScript dev reviewing for security bugs"). ' +
-            'Context provides COMPLETE data (full source file, full error log — never truncated). ' +
-            'Instruction states exactly what to produce (under 50 words for best results). ' +
-            'Best for: code review, comparison, refactoring suggestions, structured analysis. ' +
-            'Expect 30-180s response time depending on max_tokens.',
+        description: 'Structured analysis via the local LLM with explicit system/context/instruction separation. ' +
+            'This 3-part format prevents context bleed and gets the best results from local models.\n\n' +
+            'USE THIS for complex tasks where prompt structure matters — it consistently outperforms ' +
+            'stuffing everything into a single message. The separation helps the local model focus.\n\n' +
+            'WHEN TO USE:\n' +
+            '• Code review — paste full source, ask for bugs/improvements\n' +
+            '• Comparison — paste two implementations, ask which is better and why\n' +
+            '• Refactoring suggestions — paste code, ask for a cleaner version\n' +
+            '• Content analysis — paste text, ask for structure/tone/issues\n' +
+            '• Any task where separating context from instruction improves clarity\n\n' +
+            'PROMPT STRUCTURE (each field has a job — keep them focused):\n' +
+            '• System: persona + constraints, under 30 words. "Expert Python developer focused on performance and correctness."\n' +
+            '• Context: COMPLETE data. Full source code, full logs, full text. NEVER truncate or summarise.\n' +
+            '• Instruction: exactly what to produce, under 50 words. Specify format: "Return a JSON array of {line, issue, fix}."\n\n' +
+            'QA: Review the output. The local model is a capable drafter — verify its analysis before acting on it.',
         inputSchema: {
             type: 'object',
             properties: {
-                system: { type: 'string', description: 'Persona. Be specific and under 30 words. Example: "Expert Node.js developer focused on error handling and edge cases."' },
-                context: { type: 'string', description: 'The COMPLETE data to analyse. Full source code, full logs, full text. NEVER truncate — the local LLM fills gaps with plausible hallucinations.' },
-                instruction: { type: 'string', description: 'What to produce. Under 50 words. Specify format: "List 3 bugs as bullet points" or "Return a JSON array of {line, issue, fix}".' },
-                temperature: { type: 'number', description: '0.1 for bugs/review, 0.3 for analysis (default), 0.5 for suggestions.' },
-                max_tokens: { type: 'number', description: 'Match to expected output. 200 for bullets, 400 for detailed review, 600 for code generation.' },
+                system: {
+                    type: 'string',
+                    description: 'Persona. Be specific: "Expert Node.js developer focused on error handling and edge cases."',
+                },
+                context: {
+                    type: 'string',
+                    description: 'The COMPLETE data to analyse. Full source code, full logs, full text. NEVER truncate.',
+                },
+                instruction: {
+                    type: 'string',
+                    description: 'What to produce. Specify format: "List 3 bugs as bullet points" or "Return a JSON array of {line, issue, fix}".',
+                },
+                temperature: {
+                    type: 'number',
+                    description: '0.1 for bugs/review, 0.3 for analysis (default), 0.5 for suggestions.',
+                },
+                max_tokens: {
+                    type: 'number',
+                    description: 'Max response tokens. Default 2048.',
+                },
             },
             required: ['instruction'],
         },
     },
     {
         name: 'code_task',
-        description: 'Purpose-built for code analysis tasks. Wraps the local LLM with an optimised code-review system prompt. ' +
-            'Provide COMPLETE source code and a specific task — returns analysis in ~30-180s. ' +
-            'Ideal for parallel execution: delegate to local LLM while you handle other work. ' +
-            'The local LLM excels at: explaining code, finding common bugs, suggesting improvements, comparing patterns, generating boilerplate. ' +
-            'It struggles with: subtle/adversarial bugs, multi-file reasoning, design tasks requiring integration. ' +
-            'Output capped at 500 tokens by default (override with max_tokens).',
+        description: 'Send a code analysis task to the local LLM. Wraps the request with an optimised code-review system prompt.\n\n' +
+            'This is the fastest way to offload code-specific work. Temperature is locked to 0.2 for ' +
+            'focused, deterministic output. The system prompt is pre-configured for code review.\n\n' +
+            'WHEN TO USE:\n' +
+            '• Explain what a function/class does\n' +
+            '• Find bugs or suggest improvements\n' +
+            '• Generate unit tests or type definitions for existing code\n' +
+            '• Add error handling, logging, or validation\n' +
+            '• Convert between languages or patterns\n\n' +
+            'GETTING BEST RESULTS:\n' +
+            '• Provide COMPLETE source code — the local LLM cannot read files.\n' +
+            '• Include imports and type definitions so the model has full context.\n' +
+            '• Be specific in the task: "Write 3 Jest tests for the error paths in fetchUser" beats "Write tests".\n' +
+            '• Set the language field — it shapes the system prompt and improves accuracy.\n\n' +
+            'QA: Always verify generated code compiles, handles edge cases, and follows project conventions.',
         inputSchema: {
             type: 'object',
             properties: {
-                code: { type: 'string', description: 'COMPLETE source code. Never truncate. Include imports and full function bodies.' },
-                task: { type: 'string', description: 'What to do. Be specific and concise: "Find bugs", "Explain this function", "Add error handling to fetchData", "Compare these two approaches".' },
-                language: { type: 'string', description: 'Programming language for context: "typescript", "python", "rust", etc.' },
-                max_tokens: { type: 'number', description: 'Expected output size. Default 500. Use 200 for quick answers, 800 for code generation.' },
+                code: {
+                    type: 'string',
+                    description: 'COMPLETE source code. Never truncate. Include imports and full function bodies.',
+                },
+                task: {
+                    type: 'string',
+                    description: 'What to do: "Find bugs", "Explain this", "Add error handling to fetchData", "Write tests".',
+                },
+                language: {
+                    type: 'string',
+                    description: 'Programming language: "typescript", "python", "rust", etc.',
+                },
+                max_tokens: {
+                    type: 'number',
+                    description: 'Max response tokens. Default 2048.',
+                },
             },
             required: ['code', 'task'],
         },
     },
     {
-        name: 'list_models',
-        description: 'List models currently loaded in LM Studio. Use to verify which model will handle delegated tasks.',
+        name: 'discover',
+        description: 'Check whether the local LLM is online and what model is loaded. Returns model name, context window size, ' +
+            'response latency, and cumulative session stats (tokens offloaded so far). ' +
+            'Call this if you are unsure whether the local LLM is available before delegating work. ' +
+            'Fast — typically responds in under 1 second, or returns an offline status within 5 seconds if the host is unreachable.',
         inputSchema: { type: 'object', properties: {} },
     },
     {
-        name: 'health_check',
-        description: 'Check connectivity and latency to the local LM Studio instance. Run before delegating time-sensitive tasks.',
+        name: 'list_models',
+        description: 'List all models currently loaded in the local LLM server, with context window sizes. ' +
+            'Use discover instead for a quick availability check.',
         inputSchema: { type: 'object', properties: {} },
     },
 ];
 // ── MCP Server ───────────────────────────────────────────────────────
-const server = new Server({ name: 'houtini-lm', version: '2.1.0' }, { capabilities: { tools: {} } });
+const server = new Server({ name: 'houtini-lm', version: '2.4.1' }, { capabilities: { tools: {} } });
 server.setRequestHandler(ListToolsRequestSchema, async () => ({ tools: TOOLS }));
 server.setRequestHandler(CallToolRequestSchema, async (request) => {
     const { name, arguments: args } = request.params;
@@ -133,15 +359,12 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
                 if (system)
                     messages.push({ role: 'system', content: system });
                 messages.push({ role: 'user', content: message });
-                const resp = await chatCompletion(messages, {
+                const resp = await chatCompletionStreaming(messages, {
                     temperature,
                     maxTokens: max_tokens,
                 });
-                const reply = resp.choices[0]?.message?.content ?? '';
-                const usage = resp.usage
-                    ? `\n\n---\nModel: ${resp.model} | Tokens: ${resp.usage.prompt_tokens}→${resp.usage.completion_tokens}`
-                    : '';
-                return { content: [{ type: 'text', text: reply + usage }] };
+                const footer = formatFooter(resp);
+                return { content: [{ type: 'text', text: resp.content + footer }] };
             }
             case 'custom_prompt': {
                 const { system, context, instruction, temperature, max_tokens } = args;
@@ -152,12 +375,13 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
                 if (context)
                     userContent = `Context:\n${context}\n\nInstruction:\n${instruction}`;
                 messages.push({ role: 'user', content: userContent });
-                const resp = await chatCompletion(messages, {
+                const resp = await chatCompletionStreaming(messages, {
                     temperature,
                     maxTokens: max_tokens,
                 });
+                const footer = formatFooter(resp);
                 return {
-                    content: [{ type: 'text', text: resp.choices[0]?.message?.content ?? '' }],
+                    content: [{ type: 'text', text: resp.content + footer }],
                 };
             }
             case 'code_task': {
@@ -173,40 +397,74 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
                         content: `Task: ${task}\n\n\`\`\`${lang}\n${code}\n\`\`\``,
                     },
                 ];
-                const codeResp = await chatCompletion(codeMessages, {
+                const codeResp = await chatCompletionStreaming(codeMessages, {
                     temperature: 0.2,
-                    maxTokens: codeMaxTokens ?? 500,
+                    maxTokens: codeMaxTokens ?? DEFAULT_MAX_TOKENS,
                 });
-                const codeReply = codeResp.choices[0]?.message?.content ?? '';
-                const codeUsage = codeResp.usage
-                    ? `\n\n---\nModel: ${codeResp.model} | Tokens: ${codeResp.usage.prompt_tokens}→${codeResp.usage.completion_tokens} | ${lang}`
-                    : '';
-                return { content: [{ type: 'text', text: codeReply + codeUsage }] };
+                const codeFooter = formatFooter(codeResp, lang);
+                return { content: [{ type: 'text', text: codeResp.content + codeFooter }] };
             }
-            case 'list_models': {
-                const models = await listModels();
+            case 'discover': {
+                const start = Date.now();
+                let models;
+                try {
+                    models = await listModelsRaw();
+                }
+                catch (err) {
+                    const ms = Date.now() - start;
+                    const reason = err instanceof Error && err.name === 'AbortError'
+                        ? `Host unreachable (timed out after ${ms}ms)`
+                        : `Connection failed: ${err instanceof Error ? err.message : String(err)}`;
+                    return {
+                        content: [{
+                                type: 'text',
+                                text: `Status: OFFLINE\nEndpoint: ${LM_BASE_URL}\n${reason}\n\nThe local LLM is not available right now. Do not attempt to delegate tasks to it.`,
+                            }],
+                    };
+                }
+                const ms = Date.now() - start;
+                if (models.length === 0) {
+                    return {
+                        content: [{
+                                type: 'text',
+                                text: `Status: ONLINE (no model loaded)\nEndpoint: ${LM_BASE_URL}\nLatency: ${ms}ms\n\nThe server is running but no model is loaded. Ask the user to load a model in LM Studio.`,
+                            }],
+                    };
+                }
+                const lines = models.map((m) => {
+                    const ctx = getContextLength(m);
+                    return `  • ${m.id} (context: ${ctx.toLocaleString()} tokens)`;
+                });
+                const primary = models[0];
+                const ctx = getContextLength(primary);
+                const sessionStats = session.calls > 0
+                    ? `\nSession stats: ${(session.promptTokens + session.completionTokens).toLocaleString()} tokens offloaded across ${session.calls} call${session.calls === 1 ? '' : 's'}`
+                    : '\nSession stats: no calls yet — delegate tasks to start saving tokens';
                 return {
-                    content: [
-                        {
+                    content: [{
                             type: 'text',
-                            text: models.length
-                                ? `Loaded models:\n${models.map((m) => `  • ${m}`).join('\n')}`
-                                : 'No models currently loaded.',
-                        },
-                    ],
+                            text: `Status: ONLINE\n` +
+                                `Endpoint: ${LM_BASE_URL}\n` +
+                                `Latency: ${ms}ms\n` +
+                                `Model: ${primary.id}\n` +
+                                `Context window: ${ctx.toLocaleString()} tokens\n` +
+                                `\nLoaded models:\n${lines.join('\n')}` +
+                                `${sessionStats}\n\n` +
+                                `The local LLM is available. You can delegate tasks using chat, custom_prompt, or code_task.`,
+                        }],
                 };
             }
-            case 'health_check': {
-                const start = Date.now();
-                const models = await listModels();
-                const ms = Date.now() - start;
+            case 'list_models': {
+                const models = await listModelsRaw();
+                if (!models.length) {
+                    return { content: [{ type: 'text', text: 'No models currently loaded.' }] };
+                }
+                const lines = models.map((m) => {
+                    const ctx = getContextLength(m);
+                    return `  • ${m.id}${ctx ? ` (context: ${ctx.toLocaleString()} tokens)` : ''}`;
+                });
                 return {
-                    content: [
-                        {
-                            type: 'text',
-                            text: `Connected to ${LM_BASE_URL} (${ms}ms)\nAuth: ${LM_PASSWORD ? 'enabled' : 'none'}\nModels loaded: ${models.length}${models.length ? '\n' + models.join(', ') : ''}`,
-                        },
-                    ],
+                    content: [{ type: 'text', text: `Loaded models:\n${lines.join('\n')}` }],
                 };
             }
             default:

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";AACA;;;;;GAKG;AAEH,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AACnE,OAAO,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AACjF,OAAO,EACL,qBAAqB,EACrB,sBAAsB,GACvB,MAAM,oCAAoC,CAAC;AAE5C,MAAM,WAAW,GAAG,OAAO,CAAC,GAAG,CAAC,aAAa,IAAI,uBAAuB,CAAC;AACzE,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,EAAE,CAAC;AACnD,MAAM,WAAW,GAAG,OAAO,CAAC,GAAG,CAAC,kBAAkB,IAAI,EAAE,CAAC;AACzD,MAAM,kBAAkB,GAAG,IAAI,CAAC;AAChC,MAAM,mBAAmB,GAAG,GAAG,CAAC;AAEhC,SAAS,UAAU;IACjB,MAAM,CAAC,GAA2B,EAAE,cAAc,EAAE,kBAAkB,EAAE,CAAC;IACzE,IAAI,WAAW;QAAE,CAAC,CAAC,eAAe,CAAC,GAAG,UAAU,WAAW,EAAE,CAAC;IAC9D,OAAO,CAAC,CAAC;AACX,CAAC;AAmBD,KAAK,UAAU,cAAc,CAC3B,QAAuB,EACvB,UAAwE,EAAE;IAE1E,MAAM,IAAI,GAA4B;QACpC,QAAQ;QACR,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,mBAAmB;QACvD,UAAU,EAAE,OAAO,CAAC,SAAS,IAAI,kBAAkB;QACnD,MAAM,EAAE,KAAK;KACd,CAAC;IACF,IAAI,OAAO,CAAC,KAAK,IAAI,QAAQ,EAAE,CAAC;QAC9B,IAAI,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,QAAQ,CAAC;IACzC,CAAC;IAED,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,WAAW,sBAAsB,EAAE;QAC5D,MAAM,EAAE,MAAM;QACd,OAAO,EAAE,UAAU,EAAE;QACrB,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;KAC3B,CAAC,CAAC;IAEH,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;QACZ,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,CAAC;QAC9C,MAAM,IAAI,KAAK,CAAC,uBAAuB,GAAG,CAAC,MAAM,KAAK,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IAED,OAAO,GAAG,CAAC,IAAI,EAAqC,CAAC;AACvD,CAAC;AAED,KAAK,UAAU,UAAU;IACvB,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,WAAW,YAAY,EAAE,EAAE,OAAO,EAAE,UAAU,EAAE,EAAE,CAAC,CAAC;IAC/E,IAAI,CAAC,GAAG,CAAC,EAAE;QAAE,MAAM,IAAI,KAAK,CAAC,0BAA0B,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;IACrE,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAoC,CAAC;IACnE,OAAO,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;AACpC,CAAC;AAED,wEAAwE;AAExE,MAAM,KAAK,GAAG;IACZ;QACE,IAAI,EAAE,MAAM;QACZ,WAAW,EACT,sEAAsE;YACtE,sGAAsG;YACtG,gFAAgF;YAChF,mHAAmH;YACnH,4IAA4I;YAC5I,4GAA4G;YAC5G,gLAAgL;QAClL,WAAW,EAAE;YACX,IAAI,EAAE,QAAiB;YACvB,UAAU,EAAE;gBACV,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,wGAAwG,EAAE;gBAClJ,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,4IAA4I,EAAE;gBACrL,WAAW,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,gHAAgH,EAAE;gBAC9J,UAAU,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,oJAAoJ,EAAE;aAClM;YACD,QAAQ,EAAE,CAAC,SAAS,CAAC;SACtB;KACF;IACD;QACE,IAAI,EAAE,eAAe;QACrB,WAAW,EACT,4FAA4F;YAC5F,sGAAsG;YACtG,0FAA0F;YAC1F,uFAAuF;YACvF,gFAAgF;YAChF,mFAAmF;YACnF,uDAAuD;QACzD,WAAW,EAAE;YACX,IAAI,EAAE,QAAiB;YACvB,UAAU,EAAE;gBACV,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,wHAAwH,EAAE;gBACjK,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,gJAAgJ,EAAE;gBAC1L,WAAW,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,iIAAiI,EAAE;gBAC/K,WAAW,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,uEAAuE,EAAE;gBACrH,UAAU,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,8FAA8F,EAAE;aAC5I;YACD,QAAQ,EAAE,CAAC,aAAa,CAAC;SAC1B;KACF;IACD;QACE,IAAI,EAAE,WAAW;QACjB,WAAW,EACT,0GAA0G;YAC1G,mFAAmF;YACnF,mFAAmF;YACnF,sIAAsI;YACtI,wGAAwG;YACxG,oEAAoE;QACtE,WAAW,EAAE;YACX,IAAI,EAAE,QAAiB;YACvB,UAAU,EAAE;gBACV,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,iFAAiF,EAAE;gBACxH,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,+IAA+I,EAAE;gBACtL,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,wEAAwE,EAAE;gBACnH,UAAU,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,wFAAwF,EAAE;aACtI;YACD,QAAQ,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC;SAC3B;KACF;IACD;QACE,IAAI,EAAE,aAAa;QACnB,WAAW,EAAE,mGAAmG;QAChH,WAAW,EAAE,EAAE,IAAI,EAAE,QAAiB,EAAE,UAAU,EAAE,EAAE,EAAE;KACzD;IACD;QACE,IAAI,EAAE,cAAc;QACpB,WAAW,EAAE,6GAA6G;QAC1H,WAAW,EAAE,EAAE,IAAI,EAAE,QAAiB,EAAE,UAAU,EAAE,EAAE,EAAE;KACzD;CACF,CAAC;AAEF,wEAAwE;AAExE,MAAM,MAAM,GAAG,IAAI,MAAM,CACvB,EAAE,IAAI,EAAE,YAAY,EAAE,OAAO,EAAE,OAAO,EAAE,EACxC,EAAE,YAAY,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE,CAChC,CAAC;AAEF,MAAM,CAAC,iBAAiB,CAAC,sBAAsB,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;AAEjF,MAAM,CAAC,iBAAiB,CAAC,qBAAqB,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;IAChE,MAAM,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC;IAEjD,IAAI,CAAC;QACH,QAAQ,IAAI,EAAE,CAAC;YACb,KAAK,MAAM,CAAC,CAAC,CAAC;gBACZ,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,IAKpD,CAAC;gBACF,MAAM,QAAQ,GAAkB,EAAE,CAAC;gBACnC,IAAI,MAAM;oBAAE,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gBAC/D,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC,CAAC;gBAElD,MAAM,IAAI,GAAG,MAAM,cAAc,CAAC,QAAQ,EAAE;oBAC1C,WAAW;oBACX,SAAS,EAAE,UAAU;iBACtB,CAAC,CAAC;gBAEH,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,IAAI,EAAE,CAAC;gBACtD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK;oBACtB,CAAC,CAAC,mBAAmB,IAAI,CAAC,KAAK,cAAc,IAAI,CAAC,KAAK,CAAC,aAAa,IAAI,IAAI,CAAC,KAAK,CAAC,iBAAiB,EAAE;oBACvG,CAAC,CAAC,EAAE,CAAC;gBAEP,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,KAAK,GAAG,KAAK,EAAE,CAAC,EAAE,CAAC;YAC9D,CAAC;YAED,KAAK,eAAe,CAAC,CAAC,CAAC;gBACrB,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,IAMjE,CAAC;gBAEF,MAAM,QAAQ,GAAkB,EAAE,CAAC;gBACnC,IAAI,MAAM;oBAAE,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gBAE/D,IAAI,WAAW,GAAG,WAAW,CAAC;gBAC9B,IAAI,OAAO;oBAAE,WAAW,GAAG,aAAa,OAAO,qBAAqB,WAAW,EAAE,CAAC;gBAClF,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC;gBAEtD,MAAM,IAAI,GAAG,MAAM,cAAc,CAAC,QAAQ,EAAE;oBAC1C,WAAW;oBACX,SAAS,EAAE,UAAU;iBACtB,CAAC,CAAC;gBAEH,OAAO;oBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,IAAI,EAAE,EAAE,CAAC;iBAC3E,CAAC;YACJ,CAAC;YAED,KAAK,WAAW,CAAC,CAAC,CAAC;gBACjB,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,aAAa,EAAE,GAAG,IAK3D,CAAC;gBAEF,MAAM,IAAI,GAAG,QAAQ,IAAI,SAAS,CAAC;gBACnC,MAAM,YAAY,GAAkB;oBAClC;wBACE,IAAI,EAAE,QAAQ;wBACd,OAAO,EAAE,UAAU,IAAI,qJAAqJ;qBAC7K;oBACD;wBACE,IAAI,EAAE,MAAM;wBACZ,OAAO,EAAE,SAAS,IAAI,aAAa,IAAI,KAAK,IAAI,UAAU;qBAC3D;iBACF,CAAC;gBAEF,MAAM,QAAQ,GAAG,MAAM,cAAc,CAAC,YAAY,EAAE;oBAClD,WAAW,EAAE,GAAG;oBAChB,SAAS,EAAE,aAAa,IAAI,GAAG;iBAChC,CAAC,CAAC;gBAEH,MAAM,SAAS,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,IAAI,EAAE,CAAC;gBAC9D,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK;oBAC9B,CAAC,CAAC,mBAAmB,QAAQ,CAAC,KAAK,cAAc,QAAQ,CAAC,KAAK,CAAC,aAAa,IAAI,QAAQ,CAAC,KAAK,CAAC,iBAAiB,MAAM,IAAI,EAAE;oBAC7H,CAAC,CAAC,EAAE,CAAC;gBAEP,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,SAAS,GAAG,SAAS,EAAE,CAAC,EAAE,CAAC;YACtE,CAAC;YAED,KAAK,aAAa,CAAC,CAAC,CAAC;gBACnB,MAAM,MAAM,GAAG,MAAM,UAAU,EAAE,CAAC;gBAClC,OAAO;oBACL,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EAAE,MAAM,CAAC,MAAM;gCACjB,CAAC,CAAC,mBAAmB,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;gCAC/D,CAAC,CAAC,6BAA6B;yBAClC;qBACF;iBACF,CAAC;YACJ,CAAC;YAED,KAAK,cAAc,CAAC,CAAC,CAAC;gBACpB,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;gBACzB,MAAM,MAAM,GAAG,MAAM,UAAU,EAAE,CAAC;gBAClC,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;gBAC9B,OAAO;oBACL,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EAAE,gBAAgB,WAAW,KAAK,EAAE,cAAc,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,oBAAoB,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE;yBAC1K;qBACF;iBACF,CAAC;YACJ,CAAC;YAED;gBACE,MAAM,IAAI,KAAK,CAAC,iBAAiB,IAAI,EAAE,CAAC,CAAC;QAC7C,CAAC;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO;YACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;YACrG,OAAO,EAAE,IAAI;SACd,CAAC;IACJ,CAAC;AACH,CAAC,CAAC,CAAC;AAEH,KAAK,UAAU,IAAI;IACjB,MAAM,SAAS,GAAG,IAAI,oBAAoB,EAAE,CAAC;IAC7C,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;IAChC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,8BAA8B,WAAW,KAAK,CAAC,CAAC;AACvE,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE;IACrB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,gBAAgB,KAAK,IAAI,CAAC,CAAC;IAChD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";AACA;;;;;GAKG;AAEH,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AACnE,OAAO,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AACjF,OAAO,EACL,qBAAqB,EACrB,sBAAsB,GACvB,MAAM,oCAAoC,CAAC;AAE5C,MAAM,WAAW,GAAG,OAAO,CAAC,GAAG,CAAC,aAAa,IAAI,uBAAuB,CAAC;AACzE,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,EAAE,CAAC;AACnD,MAAM,WAAW,GAAG,OAAO,CAAC,GAAG,CAAC,kBAAkB,IAAI,EAAE,CAAC;AACzD,MAAM,kBAAkB,GAAG,IAAI,CAAC;AAChC,MAAM,mBAAmB,GAAG,GAAG,CAAC;AAChC,MAAM,kBAAkB,GAAG,IAAI,CAAC;AAChC,MAAM,4BAA4B,GAAG,MAAM,CAAC,CAAC,yCAAyC;AACtF,MAAM,eAAe,GAAG,MAAM,CAAC,CAAc,qDAAqD;AAClG,MAAM,qBAAqB,GAAG,MAAM,CAAC,CAAQ,kCAAkC;AAC/E,MAAM,uBAAuB,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,QAAQ,EAAE,EAAE,CAAC,CAAC;AAExF,wEAAwE;AACxE,uEAAuE;AACvE,uEAAuE;AACvE,4DAA4D;AAE5D,MAAM,OAAO,GAAG;IACd,KAAK,EAAE,CAAC;IACR,YAAY,EAAE,CAAC;IACf,gBAAgB,EAAE,CAAC;CACpB,CAAC;AAEF,SAAS,WAAW,CAAC,KAA4D;IAC/E,OAAO,CAAC,KAAK,EAAE,CAAC;IAChB,IAAI,KAAK,EAAE,CAAC;QACV,OAAO,CAAC,YAAY,IAAI,KAAK,CAAC,aAAa,CAAC;QAC5C,OAAO,CAAC,gBAAgB,IAAI,KAAK,CAAC,iBAAiB,CAAC;IACtD,CAAC;AACH,CAAC;AAED,SAAS,cAAc;IACrB,MAAM,KAAK,GAAG,OAAO,CAAC,YAAY,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAC9D,IAAI,OAAO,CAAC,KAAK,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACnC,OAAO,YAAY,KAAK,CAAC,cAAc,EAAE,4BAA4B,OAAO,CAAC,KAAK,QAAQ,OAAO,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC;AAC7H,CAAC;AAED,SAAS,UAAU;IACjB,MAAM,CAAC,GAA2B,EAAE,cAAc,EAAE,kBAAkB,EAAE,CAAC;IACzE,IAAI,WAAW;QAAE,CAAC,CAAC,eAAe,CAAC,GAAG,UAAU,WAAW,EAAE,CAAC;IAC9D,OAAO,CAAC,CAAC;AACX,CAAC;AAyBD;;GAEG;AACH,KAAK,UAAU,gBAAgB,CAC7B,GAAW,EACX,OAAoB,EACpB,YAAoB,kBAAkB;IAEtC,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;IAC9D,IAAI,CAAC;QACH,OAAO,MAAM,KAAK,CAAC,GAAG,EAAE,EAAE,GAAG,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;IACrE,CAAC;YAAS,CAAC;QACT,YAAY,CAAC,KAAK,CAAC,CAAC;IACtB,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,KAAK,UAAU,SAAS,CACtB,MAA+C,EAC/C,SAAiB;IAEjB,IAAI,KAAoC,CAAC;IACzC,MAAM,OAAO,GAAG,IAAI,OAAO,CAAY,CAAC,OAAO,EAAE,EAAE;QACjD,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,SAAS,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;IACH,IAAI,CAAC;QACH,OAAO,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,OAAO,CAAC,CAAC,CAAC;IACtD,CAAC;YAAS,CAAC;QACT,YAAY,CAAC,KAAM,CAAC,CAAC;IACvB,CAAC;AACH,CAAC;AAED;;;;;;;GAOG;AACH,KAAK,UAAU,uBAAuB,CACpC,QAAuB,EACvB,UAAwE,EAAE;IAE1E,MAAM,IAAI,GAA4B;QACpC,QAAQ;QACR,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,mBAAmB;QACvD,UAAU,EAAE,OAAO,CAAC,SAAS,IAAI,kBAAkB;QACnD,MAAM,EAAE,IAAI;KACb,CAAC;IACF,IAAI,OAAO,CAAC,KAAK,IAAI,QAAQ,EAAE,CAAC;QAC9B,IAAI,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,QAAQ,CAAC;IACzC,CAAC;IAED,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAE7B,MAAM,GAAG,GAAG,MAAM,gBAAgB,CAChC,GAAG,WAAW,sBAAsB,EACpC,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,EAAE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,EACrE,4BAA4B,CAC7B,CAAC;IAEF,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;QACZ,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,CAAC;QAC9C,MAAM,IAAI,KAAK,CAAC,uBAAuB,GAAG,CAAC,MAAM,KAAK,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IAED,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC;QACd,MAAM,IAAI,KAAK,CAAC,6DAA6D,CAAC,CAAC;IACjF,CAAC;IAED,MAAM,MAAM,GAAG,GAAG,CAAC,IAAI,CAAC,SAAS,EAAE,CAAC;IACpC,MAAM,OAAO,GAAG,IAAI,WAAW,EAAE,CAAC;IAClC,IAAI,OAAO,GAAG,EAAE,CAAC;IACjB,IAAI,KAAK,GAAG,EAAE,CAAC;IACf,IAAI,KAA+B,CAAC;IACpC,IAAI,YAAY,GAAG,EAAE,CAAC;IACtB,IAAI,SAAS,GAAG,KAAK,CAAC;IACtB,IAAI,MAAM,GAAG,EAAE,CAAC;IAEhB,IAAI,CAAC;QACH,OAAO,IAAI,EAAE,CAAC;YACZ,sCAAsC;YACtC,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;YACvC,IAAI,OAAO,GAAG,eAAe,EAAE,CAAC;gBAC9B,SAAS,GAAG,IAAI,CAAC;gBACjB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,gCAAgC,OAAO,iBAAiB,OAAO,CAAC,MAAM,6BAA6B,CAAC,CAAC;gBAC1H,MAAM;YACR,CAAC;YAED,2DAA2D;YAC3D,MAAM,SAAS,GAAG,eAAe,GAAG,OAAO,CAAC;YAC5C,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,qBAAqB,EAAE,SAAS,CAAC,CAAC;YAChE,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;YAErD,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;gBACzB,SAAS,GAAG,IAAI,CAAC;gBACjB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,8CAA8C,OAAO,CAAC,MAAM,6BAA6B,CAAC,CAAC;gBAChH,MAAM;YACR,CAAC;YAED,IAAI,MAAM,CAAC,IAAI;gBAAE,MAAM;YAEvB,MAAM,IAAI,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC;YAEzD,kBAAkB;YAClB,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;YACjC,MAAM,GAAG,KAAK,CAAC,GAAG,EAAE,IAAI,EAAE,CAAC,CAAC,iCAAiC;YAE7D,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;gBACzB,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;gBAC5B,IAAI,CAAC,OAAO,IAAI,OAAO,KAAK,cAAc;oBAAE,SAAS;gBACrD,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,QAAQ,CAAC;oBAAE,SAAS;gBAE5C,IAAI,CAAC;oBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;oBAC1C,IAAI,IAAI,CAAC,KAAK;wBAAE,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;oBAEnC,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC;oBACvC,IAAI,KAAK,EAAE,OAAO;wBAAE,OAAO,IAAI,KAAK,CAAC,OAAO,CAAC;oBAE7C,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,aAAa,CAAC;oBAChD,IAAI,MAAM;wBAAE,YAAY,GAAG,MAAM,CAAC;oBAElC,4DAA4D;oBAC5D,IAAI,IAAI,CAAC,KAAK;wBAAE,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;gBACrC,CAAC;gBAAC,MAAM,CAAC;oBACP,yDAAyD;gBAC3D,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;YAAS,CAAC;QACT,2DAA2D;QAC3D,MAAM,CAAC,WAAW,EAAE,CAAC;IACvB,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,YAAY,EAAE,SAAS,EAAE,CAAC;AAC5D,CAAC;AAED,KAAK,UAAU,aAAa;IAC1B,MAAM,GAAG,GAAG,MAAM,gBAAgB,CAChC,GAAG,WAAW,YAAY,EAC1B,EAAE,OAAO,EAAE,UAAU,EAAE,EAAE,CAC1B,CAAC;IACF,IAAI,CAAC,GAAG,CAAC,EAAE;QAAE,MAAM,IAAI,KAAK,CAAC,0BAA0B,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;IACrE,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAA0B,CAAC;IACzD,OAAO,IAAI,CAAC,IAAI,CAAC;AACnB,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAgB;IACxC,gFAAgF;IAChF,OAAO,KAAK,CAAC,cAAc,IAAI,KAAK,CAAC,aAAa,IAAI,uBAAuB,CAAC;AAChF,CAAC;AAED;;GAEG;AACH,SAAS,YAAY,CAAC,IAAqB,EAAE,KAAc;IACzD,sDAAsD;IACtD,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAExB,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,IAAI,CAAC,KAAK;QAAE,KAAK,CAAC,IAAI,CAAC,UAAU,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IACnD,IAAI,IAAI,CAAC,KAAK;QAAE,KAAK,CAAC,IAAI,CAAC,cAAc,IAAI,CAAC,KAAK,CAAC,aAAa,IAAI,IAAI,CAAC,KAAK,CAAC,iBAAiB,SAAS,CAAC,CAAC;IAC5G,IAAI,KAAK;QAAE,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAC7B,IAAI,IAAI,CAAC,SAAS;QAAE,KAAK,CAAC,IAAI,CAAC,6CAA6C,CAAC,CAAC;IAE9E,MAAM,WAAW,GAAG,cAAc,EAAE,CAAC;IACrC,IAAI,WAAW;QAAE,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IAEzC,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;AACjE,CAAC;AAED,wEAAwE;AAExE,MAAM,KAAK,GAAG;IACZ;QACE,IAAI,EAAE,MAAM;QACZ,WAAW,EACT,8FAA8F;YAC9F,yGAAyG;YACzG,sDAAsD;YACtD,+FAA+F;YAC/F,4FAA4F;YAC5F,mGAAmG;YACnG,yDAAyD;YACzD,kDAAkD;YAClD,mEAAmE;YACnE,gEAAgE;YAChE,sDAAsD;YACtD,qEAAqE;YACrE,oDAAoD;YACpD,iEAAiE;YACjE,uFAAuF;YACvF,8FAA8F;YAC9F,gGAAgG;YAChG,uGAAuG;YACvG,0FAA0F;YAC1F,qGAAqG;YACrG,mGAAmG;YACnG,6GAA6G;YAC7G,mGAAmG;QACrG,WAAW,EAAE;YACX,IAAI,EAAE,QAAiB;YACvB,UAAU,EAAE;gBACV,OAAO,EAAE;oBACP,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,qGAAqG;iBACnH;gBACD,MAAM,EAAE;oBACN,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,0FAA0F;iBACxG;gBACD,WAAW,EAAE;oBACX,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,8FAA8F;iBAC5G;gBACD,UAAU,EAAE;oBACV,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,6FAA6F;iBAC3G;aACF;YACD,QAAQ,EAAE,CAAC,SAAS,CAAC;SACtB;KACF;IACD;QACE,IAAI,EAAE,eAAe;QACrB,WAAW,EACT,6FAA6F;YAC7F,4FAA4F;YAC5F,0FAA0F;YAC1F,4FAA4F;YAC5F,gBAAgB;YAChB,gEAAgE;YAChE,yEAAyE;YACzE,qEAAqE;YACrE,kEAAkE;YAClE,2EAA2E;YAC3E,gEAAgE;YAChE,sHAAsH;YACtH,kGAAkG;YAClG,0HAA0H;YAC1H,wGAAwG;QAC1G,WAAW,EAAE;YACX,IAAI,EAAE,QAAiB;YACvB,UAAU,EAAE;gBACV,MAAM,EAAE;oBACN,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,4FAA4F;iBAC1G;gBACD,OAAO,EAAE;oBACP,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,uFAAuF;iBACrG;gBACD,WAAW,EAAE;oBACX,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,iHAAiH;iBAC/H;gBACD,WAAW,EAAE;oBACX,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,uEAAuE;iBACrF;gBACD,UAAU,EAAE;oBACV,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,oCAAoC;iBAClD;aACF;YACD,QAAQ,EAAE,CAAC,aAAa,CAAC;SAC1B;KACF;IACD;QACE,IAAI,EAAE,WAAW;QACjB,WAAW,EACT,gHAAgH;YAChH,0FAA0F;YAC1F,yFAAyF;YACzF,gBAAgB;YAChB,wCAAwC;YACxC,uCAAuC;YACvC,+DAA+D;YAC/D,gDAAgD;YAChD,6CAA6C;YAC7C,yBAAyB;YACzB,qEAAqE;YACrE,yEAAyE;YACzE,yGAAyG;YACzG,mFAAmF;YACnF,iGAAiG;QACnG,WAAW,EAAE;YACX,IAAI,EAAE,QAAiB;YACvB,UAAU,EAAE;gBACV,IAAI,EAAE;oBACJ,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,iFAAiF;iBAC/F;gBACD,IAAI,EAAE;oBACJ,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,4FAA4F;iBAC1G;gBACD,QAAQ,EAAE;oBACR,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,4DAA4D;iBAC1E;gBACD,UAAU,EAAE;oBACV,IAAI,EAAE,QAAQ;oBACd,WAAW,EAAE,oCAAoC;iBAClD;aACF;YACD,QAAQ,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC;SAC3B;KACF;IACD;QACE,IAAI,EAAE,UAAU;QAChB,WAAW,EACT,2GAA2G;YAC3G,4EAA4E;YAC5E,yFAAyF;YACzF,wHAAwH;QAC1H,WAAW,EAAE,EAAE,IAAI,EAAE,QAAiB,EAAE,UAAU,EAAE,EAAE,EAAE;KACzD;IACD;QACE,IAAI,EAAE,aAAa;QACnB,WAAW,EACT,uFAAuF;YACvF,sDAAsD;QACxD,WAAW,EAAE,EAAE,IAAI,EAAE,QAAiB,EAAE,UAAU,EAAE,EAAE,EAAE;KACzD;CACF,CAAC;AAEF,wEAAwE;AAExE,MAAM,MAAM,GAAG,IAAI,MAAM,CACvB,EAAE,IAAI,EAAE,YAAY,EAAE,OAAO,EAAE,OAAO,EAAE,EACxC,EAAE,YAAY,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE,CAChC,CAAC;AAEF,MAAM,CAAC,iBAAiB,CAAC,sBAAsB,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;AAEjF,MAAM,CAAC,iBAAiB,CAAC,qBAAqB,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;IAChE,MAAM,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC;IAEjD,IAAI,CAAC;QACH,QAAQ,IAAI,EAAE,CAAC;YACb,KAAK,MAAM,CAAC,CAAC,CAAC;gBACZ,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,IAKpD,CAAC;gBACF,MAAM,QAAQ,GAAkB,EAAE,CAAC;gBACnC,IAAI,MAAM;oBAAE,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gBAC/D,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC,CAAC;gBAElD,MAAM,IAAI,GAAG,MAAM,uBAAuB,CAAC,QAAQ,EAAE;oBACnD,WAAW;oBACX,SAAS,EAAE,UAAU;iBACtB,CAAC,CAAC;gBAEH,MAAM,MAAM,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;gBAClC,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,OAAO,GAAG,MAAM,EAAE,CAAC,EAAE,CAAC;YACtE,CAAC;YAED,KAAK,eAAe,CAAC,CAAC,CAAC;gBACrB,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,IAMjE,CAAC;gBAEF,MAAM,QAAQ,GAAkB,EAAE,CAAC;gBACnC,IAAI,MAAM;oBAAE,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gBAE/D,IAAI,WAAW,GAAG,WAAW,CAAC;gBAC9B,IAAI,OAAO;oBAAE,WAAW,GAAG,aAAa,OAAO,qBAAqB,WAAW,EAAE,CAAC;gBAClF,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC;gBAEtD,MAAM,IAAI,GAAG,MAAM,uBAAuB,CAAC,QAAQ,EAAE;oBACnD,WAAW;oBACX,SAAS,EAAE,UAAU;iBACtB,CAAC,CAAC;gBAEH,MAAM,MAAM,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;gBAClC,OAAO;oBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,OAAO,GAAG,MAAM,EAAE,CAAC;iBACzD,CAAC;YACJ,CAAC;YAED,KAAK,WAAW,CAAC,CAAC,CAAC;gBACjB,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,aAAa,EAAE,GAAG,IAK3D,CAAC;gBAEF,MAAM,IAAI,GAAG,QAAQ,IAAI,SAAS,CAAC;gBACnC,MAAM,YAAY,GAAkB;oBAClC;wBACE,IAAI,EAAE,QAAQ;wBACd,OAAO,EAAE,UAAU,IAAI,qJAAqJ;qBAC7K;oBACD;wBACE,IAAI,EAAE,MAAM;wBACZ,OAAO,EAAE,SAAS,IAAI,aAAa,IAAI,KAAK,IAAI,UAAU;qBAC3D;iBACF,CAAC;gBAEF,MAAM,QAAQ,GAAG,MAAM,uBAAuB,CAAC,YAAY,EAAE;oBAC3D,WAAW,EAAE,GAAG;oBAChB,SAAS,EAAE,aAAa,IAAI,kBAAkB;iBAC/C,CAAC,CAAC;gBAEH,MAAM,UAAU,GAAG,YAAY,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;gBAChD,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,CAAC,OAAO,GAAG,UAAU,EAAE,CAAC,EAAE,CAAC;YAC9E,CAAC;YAED,KAAK,UAAU,CAAC,CAAC,CAAC;gBAChB,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;gBACzB,IAAI,MAAmB,CAAC;gBACxB,IAAI,CAAC;oBACH,MAAM,GAAG,MAAM,aAAa,EAAE,CAAC;gBACjC,CAAC;gBAAC,OAAO,GAAG,EAAE,CAAC;oBACb,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;oBAC9B,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,IAAI,GAAG,CAAC,IAAI,KAAK,YAAY;wBAC9D,CAAC,CAAC,qCAAqC,EAAE,KAAK;wBAC9C,CAAC,CAAC,sBAAsB,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC;oBAC7E,OAAO;wBACL,OAAO,EAAE,CAAC;gCACR,IAAI,EAAE,MAAM;gCACZ,IAAI,EAAE,8BAA8B,WAAW,KAAK,MAAM,uFAAuF;6BAClJ,CAAC;qBACH,CAAC;gBACJ,CAAC;gBACD,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;gBAE9B,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;oBACxB,OAAO;wBACL,OAAO,EAAE,CAAC;gCACR,IAAI,EAAE,MAAM;gCACZ,IAAI,EAAE,+CAA+C,WAAW,cAAc,EAAE,gGAAgG;6BACjL,CAAC;qBACH,CAAC;gBACJ,CAAC;gBAED,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;oBAC7B,MAAM,GAAG,GAAG,gBAAgB,CAAC,CAAC,CAAC,CAAC;oBAChC,OAAO,OAAO,CAAC,CAAC,EAAE,cAAc,GAAG,CAAC,cAAc,EAAE,UAAU,CAAC;gBACjE,CAAC,CAAC,CAAC;gBAEH,MAAM,OAAO,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;gBAC1B,MAAM,GAAG,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;gBAEtC,MAAM,YAAY,GAAG,OAAO,CAAC,KAAK,GAAG,CAAC;oBACpC,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,YAAY,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAAC,cAAc,EAAE,4BAA4B,OAAO,CAAC,KAAK,QAAQ,OAAO,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,EAAE;oBACzK,CAAC,CAAC,uEAAuE,CAAC;gBAE5E,OAAO;oBACL,OAAO,EAAE,CAAC;4BACR,IAAI,EAAE,MAAM;4BACZ,IAAI,EACF,kBAAkB;gCAClB,aAAa,WAAW,IAAI;gCAC5B,YAAY,EAAE,MAAM;gCACpB,UAAU,OAAO,CAAC,EAAE,IAAI;gCACxB,mBAAmB,GAAG,CAAC,cAAc,EAAE,WAAW;gCAClD,qBAAqB,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;gCACvC,GAAG,YAAY,MAAM;gCACrB,6FAA6F;yBAChG,CAAC;iBACH,CAAC;YACJ,CAAC;YAED,KAAK,aAAa,CAAC,CAAC,CAAC;gBACnB,MAAM,MAAM,GAAG,MAAM,aAAa,EAAE,CAAC;gBACrC,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;oBACnB,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,6BAA6B,EAAE,CAAC,EAAE,CAAC;gBAC9E,CAAC;gBACD,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;oBAC7B,MAAM,GAAG,GAAG,gBAAgB,CAAC,CAAC,CAAC,CAAC;oBAChC,OAAO,OAAO,CAAC,CAAC,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,cAAc,GAAG,CAAC,cAAc,EAAE,UAAU,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;gBACjF,CAAC,CAAC,CAAC;gBACH,OAAO;oBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,mBAAmB,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC;iBACzE,CAAC;YACJ,CAAC;YAED;gBACE,MAAM,IAAI,KAAK,CAAC,iBAAiB,IAAI,EAAE,CAAC,CAAC;QAC7C,CAAC;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO;YACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;YACrG,OAAO,EAAE,IAAI;SACd,CAAC;IACJ,CAAC;AACH,CAAC,CAAC,CAAC;AAEH,KAAK,UAAU,IAAI;IACjB,MAAM,SAAS,GAAG,IAAI,oBAAoB,EAAE,CAAC;IAC7C,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;IAChC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,8BAA8B,WAAW,KAAK,CAAC,CAAC;AACvE,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE;IACrB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,gBAAgB,KAAK,IAAI,CAAC,CAAC;IAChD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@houtini/lm",
-  "version": "2.1.0",
+  "version": "2.4.1",
   "type": "module",
   "description": "MCP server for local LLMs — connects to LM Studio or any OpenAI-compatible endpoint",
   "mcpName": "io.github.houtini-ai/lm",