theroundtaible 0.3.10 → 0.4.0

package/README.md

@@ -66,11 +66,12 @@ roundtable --version
 ### Prerequisites
 
 - Node.js 20+
-- At least one AI CLI tool installed:
+- At least one AI tool:
   - [Claude Code](https://docs.anthropic.com/en/docs/claude-code) — `claude` command
   - [Gemini CLI](https://github.com/google-gemini/gemini-cli) — `gemini` command
   - [Codex CLI](https://github.com/openai/codex) — `codex` command
-  - Or an OpenAI API key (`OPENAI_API_KEY` env var) as fallback
+  - OpenAI API key (`OPENAI_API_KEY` env var) as fallback
+  - [Ollama](https://ollama.com/) or [LM Studio](https://lmstudio.ai/) for local models
 
 ## Quick start
 
@@ -153,6 +154,7 @@ After `roundtable init`, your `.roundtable/config.json` controls everything:
 | `gemini-cli` | `gemini -p "prompt"` | Gemini Advanced | `gemini-api` |
 | `openai-cli` | `codex "prompt"` | ChatGPT Pro | `openai-api` |
 | `openai-api` | OpenAI REST API | No (API key) | — |
+| `local-llm` | OpenAI-compat or Ollama native | No (runs locally) | — |
 
 ## Code-Red mode
 
@@ -235,11 +237,59 @@ See [architecture-docs.md](architecture-docs.md) for the full technical architec
 - [x] Chronicle (persistent decision memory)
 - [x] Implementation Manifest (tracks what's been built)
 - [x] Scoped Apply (knights declare files, apply enforces scope)
+- [x] Local LLM support (Ollama, LM Studio)
 - [ ] VS Code extension
 - [ ] Web dashboard for session visualization
 - [ ] More adapters (DeepSeek, Llama, Mistral)
 - [ ] CI/CD integration (GitHub Actions)
 
+## Local LLMs
+
+TheRoundtAIble supports local models through [Ollama](https://ollama.com/) and [LM Studio](https://lmstudio.ai/). Run `roundtable init` and any running local server will be auto-detected.
+
+### Which platform should I use?
+
+**We recommend Ollama.** Here's why:
+
+| | Ollama | LM Studio |
+|---|---|---|
+| Context window | Auto-detected, set programmatically | Manual setup required |
+| Headless / CI | Yes (`ollama serve`) | No (GUI required) |
+| Model switching | Automatic via API | Manual in GUI |
+| Multi-model | Run multiple models concurrently | One model at a time |
+| Setup | `ollama pull model-name` | Download through GUI |
+
+Ollama gives TheRoundtAIble full programmatic control — context window size is detected automatically and allocated dynamically per prompt. LM Studio requires you to manually configure Context Length and Response Limit in the GUI.
+
+### Hardware requirements
+
+Local models run on your GPU. Bigger models = better discussion quality, but more VRAM:
+
+| Model size | VRAM needed | Discussion quality |
+|---|---|---|
+| 7B parameters | ~4-6 GB | Basic responses, limited reasoning |
+| 14B parameters | ~8-10 GB | Can participate, but may repeat itself |
+| 30B+ parameters | ~16-24 GB | Meaningful contributions, can hold multi-round debates |
+| 70B+ parameters | ~32-48 GB | Comparable to cloud models |
+
+**Our honest take:** Models under 30B struggle with multi-round discussions. They tend to repeat themselves, ignore other knights' arguments, and miss the collaborative spirit (no roasting!). Cloud knights (Claude, Gemini, GPT) handle 100K+ tokens and produce richer debates. Local models shine when you want privacy or offline access — but for best results, mix them with at least one cloud knight.
+
+### LM Studio setup
+
+If you choose LM Studio, you **must** manually adjust these settings (Developer tab > Model Settings):
+
+- **Context Length:** increase to at least 16384 (default 4096 is too small)
+- **Response Limit:** uncheck the limit, or set to 4096+
+
+Higher context = more VRAM and slower responses. Find the sweet spot for your GPU.
+
+### How it works under the hood
+
+- **Ollama:** Uses the native `/api/chat` endpoint with dynamic `num_ctx` — only allocates as much context as the prompt needs, saving GPU memory
+- **LM Studio:** Uses the OpenAI-compatible `/v1/chat/completions` endpoint
+- **Auto-detection:** `roundtable init` probes `localhost:11434` (Ollama) and `localhost:1234` (LM Studio), discovers loaded models, and filters out non-chat models (embeddings, TTS, etc.)
+- **Context budgeting:** The orchestrator detects each local model's context window and adjusts the source code payload so it fits — cloud knights get the full context, local knights get a trimmed version
+
 ## Known Limitations
 
 ### `roundtable apply` — BETA, use at own risk

package/dist/adapters/base.d.ts

@@ -7,6 +7,12 @@ export declare abstract class BaseAdapter {
     abstract readonly name: string;
     abstract execute(prompt: string, timeoutMs?: number): Promise<string>;
     abstract isAvailable(): Promise<boolean>;
+    /**
+     * Max chars for source context injection. Local adapters return a budget
+     * based on their detected context window. Cloud adapters return undefined
+     * (= use default 200KB).
+     */
+    getMaxSourceChars(): number | undefined;
     parseConsensus(response: string, round: number): ConsensusBlock | null;
 }
 //# sourceMappingURL=base.d.ts.map

package/dist/adapters/base.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"base.d.ts","sourceRoot":"","sources":["../../src/adapters/base.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAElD,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AACjE,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAG3D,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,CAAC;AACvC,YAAY,EAAE,gBAAgB,EAAE,CAAC;AAEjC,8BAAsB,WAAW;IAC/B,QAAQ,CAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAE/B,QAAQ,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IAErE,QAAQ,CAAC,WAAW,IAAI,OAAO,CAAC,OAAO,CAAC;IAExC,cAAc,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,cAAc,GAAG,IAAI;CAGvE"}
+{"version":3,"file":"base.d.ts","sourceRoot":"","sources":["../../src/adapters/base.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAElD,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AACjE,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAG3D,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,CAAC;AACvC,YAAY,EAAE,gBAAgB,EAAE,CAAC;AAEjC,8BAAsB,WAAW;IAC/B,QAAQ,CAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAE/B,QAAQ,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IAErE,QAAQ,CAAC,WAAW,IAAI,OAAO,CAAC,OAAO,CAAC;IAExC;;;;OAIG;IACH,iBAAiB,IAAI,MAAM,GAAG,SAAS;IAIvC,cAAc,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,cAAc,GAAG,IAAI;CAGvE"}

package/dist/adapters/base.js

@@ -3,6 +3,14 @@ import { AdapterError, classifyError } from "../utils/errors.js";
 // Re-export for backward compatibility (orchestrator imports from here)
 export { AdapterError, classifyError };
 export class BaseAdapter {
+    /**
+     * Max chars for source context injection. Local adapters return a budget
+     * based on their detected context window. Cloud adapters return undefined
+     * (= use default 200KB).
+     */
+    getMaxSourceChars() {
+        return undefined;
+    }
     parseConsensus(response, round) {
         return parseConsensusFromResponse(response, this.name, round);
     }

package/dist/adapters/base.js.map

@@ -1 +1 @@
-{"version":3,"file":"base.js","sourceRoot":"","sources":["../../src/adapters/base.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,0BAA0B,EAAE,MAAM,iBAAiB,CAAC;AAC7D,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAGjE,wEAAwE;AACxE,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,CAAC;AAGvC,MAAM,OAAgB,WAAW;IAO/B,cAAc,CAAC,QAAgB,EAAE,KAAa;QAC5C,OAAO,0BAA0B,CAAC,QAAQ,EAAE,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IAChE,CAAC;CACF"}
+{"version":3,"file":"base.js","sourceRoot":"","sources":["../../src/adapters/base.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,0BAA0B,EAAE,MAAM,iBAAiB,CAAC;AAC7D,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAGjE,wEAAwE;AACxE,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,CAAC;AAGvC,MAAM,OAAgB,WAAW;IAO/B;;;;OAIG;IACH,iBAAiB;QACf,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,cAAc,CAAC,QAAgB,EAAE,KAAa;QAC5C,OAAO,0BAA0B,CAAC,QAAQ,EAAE,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IAChE,CAAC;CACF"}

package/dist/adapters/local-llm.d.ts

@@ -0,0 +1,48 @@
+import { BaseAdapter } from "./base.js";
+type LocalSource = "Ollama" | "LM Studio" | undefined;
+export declare class LocalLlmAdapter extends BaseAdapter {
+    readonly name: string;
+    private endpoint;
+    private model;
+    private source;
+    private defaultTimeout;
+    private detectedContextTokens;
+    constructor(endpoint: string, model: string, name: string, source?: LocalSource, timeoutMs?: number);
+    isAvailable(): Promise<boolean>;
+    /**
+     * Detect the model's context window size.
+     * - Ollama: POST /api/show → model_info["*.context_length"]
+     * - LM Studio: no reliable API detection (user sets it manually)
+     */
+    detectContextWindow(): Promise<number | undefined>;
+    /**
+     * Max source chars budget based on detected or assumed context window.
+     * - Ollama: uses detected context from /api/show
+     * - LM Studio: assumes 16384 (common default, can't detect via API)
+     * - Unknown: returns undefined (use default 200KB)
+     * Reserves 4096 tokens for response + 3000 for system prompt/rounds overhead.
+     */
+    getMaxSourceChars(): number | undefined;
+    execute(prompt: string, timeoutMs?: number): Promise<string>;
+    /**
+     * Ollama native API — uses /api/chat with dynamic num_ctx based on prompt size.
+     * Only allocates as much context as needed to save GPU memory.
+     */
+    private executeOllama;
+    /**
+     * OpenAI-compatible API — used by LM Studio and unknown sources.
+     * Parses LM Studio-specific context window errors into actionable messages.
+     */
+    private executeOpenAICompat;
+    /**
+     * Detect Ollama model context window via /api/show.
+     * Looks for context_length in model_info (e.g. "llama.context_length": 32768).
+     */
+    private detectOllamaContext;
+    /**
+     * Detect LM Studio context window errors from response body.
+     */
+    private isContextWindowError;
+}
+export {};
+//# sourceMappingURL=local-llm.d.ts.map

package/dist/adapters/local-llm.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"local-llm.d.ts","sourceRoot":"","sources":["../../src/adapters/local-llm.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,WAAW,CAAC;AAGxC,KAAK,WAAW,GAAG,QAAQ,GAAG,WAAW,GAAG,SAAS,CAAC;AAEtD,qBAAa,eAAgB,SAAQ,WAAW;IAC9C,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IAEtB,OAAO,CAAC,QAAQ,CAAS;IACzB,OAAO,CAAC,KAAK,CAAS;IACtB,OAAO,CAAC,MAAM,CAAc;IAC5B,OAAO,CAAC,cAAc,CAAS;IAC/B,OAAO,CAAC,qBAAqB,CAAqB;gBAEtC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,EAAE,SAAS,GAAE,MAAgB;IAStG,WAAW,IAAI,OAAO,CAAC,OAAO,CAAC;IAcrC;;;;OAIG;IACG,mBAAmB,IAAI,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC;IAQxD;;;;;;OAMG;IACM,iBAAiB,IAAI,MAAM,GAAG,SAAS;IAc1C,OAAO,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IAmBlE;;;OAGG;YACW,aAAa;IAmD3B;;;OAGG;YACW,mBAAmB;IAmDjC;;;OAGG;YACW,mBAAmB;IAgCjC;;OAEG;IACH,OAAO,CAAC,oBAAoB;CAS7B"}

package/dist/adapters/local-llm.js

@@ -0,0 +1,218 @@
+import { BaseAdapter } from "./base.js";
+import { classifyError } from "../utils/errors.js";
+export class LocalLlmAdapter extends BaseAdapter {
+    name;
+    endpoint;
+    model;
+    source;
+    defaultTimeout;
+    detectedContextTokens;
+    constructor(endpoint, model, name, source, timeoutMs = 120_000) {
+        super();
+        this.endpoint = endpoint.replace(/\/+$/, ""); // strip trailing slash
+        this.model = model;
+        this.name = name;
+        this.source = source;
+        this.defaultTimeout = timeoutMs;
+    }
+    async isAvailable() {
+        try {
+            const controller = new AbortController();
+            const timer = setTimeout(() => controller.abort(), 3000);
+            const response = await fetch(`${this.endpoint}/v1/models`, {
+                signal: controller.signal,
+            });
+            clearTimeout(timer);
+            return response.ok;
+        }
+        catch {
+            return false;
+        }
+    }
+    /**
+     * Detect the model's context window size.
+     * - Ollama: POST /api/show → model_info["*.context_length"]
+     * - LM Studio: no reliable API detection (user sets it manually)
+     */
+    async detectContextWindow() {
+        if (this.source === "Ollama") {
+            this.detectedContextTokens = await this.detectOllamaContext();
+        }
+        // LM Studio: no reliable API, user configures manually
+        return this.detectedContextTokens;
+    }
+    /**
+     * Max source chars budget based on detected or assumed context window.
+     * - Ollama: uses detected context from /api/show
+     * - LM Studio: assumes 16384 (common default, can't detect via API)
+     * - Unknown: returns undefined (use default 200KB)
+     * Reserves 4096 tokens for response + 3000 for system prompt/rounds overhead.
+     */
+    getMaxSourceChars() {
+        // Use detected context, or conservative default for LM Studio
+        const contextTokens = this.detectedContextTokens
+            ?? (this.source === "LM Studio" ? 16384 : undefined);
+        if (!contextTokens)
+            return undefined;
+        const responseReserve = 4096;
+        const overheadReserve = 3000; // system prompt, chronicle, discussion rounds
+        const availableTokens = Math.max(contextTokens - responseReserve - overheadReserve, 2000);
+        // ~4 chars per token (rough estimate)
+        return availableTokens * 4;
+    }
+    async execute(prompt, timeoutMs) {
+        try {
+            if (this.source === "Ollama") {
+                return await this.executeOllama(prompt, timeoutMs);
+            }
+            return await this.executeOpenAICompat(prompt, timeoutMs);
+        }
+        catch (error) {
+            // Retry once on transient "Model reloaded" error (LM Studio reloads model after settings change)
+            if (error instanceof Error && error.message.includes("Model reloaded")) {
+                await new Promise((r) => setTimeout(r, 3000));
+                if (this.source === "Ollama") {
+                    return this.executeOllama(prompt, timeoutMs);
+                }
+                return this.executeOpenAICompat(prompt, timeoutMs);
+            }
+            throw error;
+        }
+    }
+    /**
+     * Ollama native API — uses /api/chat with dynamic num_ctx based on prompt size.
+     * Only allocates as much context as needed to save GPU memory.
+     */
+    async executeOllama(prompt, timeoutMs) {
+        const timeout = timeoutMs ?? this.defaultTimeout;
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), timeout);
+        // Dynamic num_ctx: prompt tokens + response budget + safety margin
+        const estimatedPromptTokens = Math.ceil(prompt.length / 4);
+        const responsebudget = 4096;
+        const safetyMargin = 512;
+        let numCtx = estimatedPromptTokens + responsebudget + safetyMargin;
+        // Clamp to model's detected max (if known)
+        if (this.detectedContextTokens) {
+            numCtx = Math.min(numCtx, this.detectedContextTokens);
+        }
+        try {
+            const response = await fetch(`${this.endpoint}/api/chat`, {
+                method: "POST",
+                headers: { "Content-Type": "application/json" },
+                body: JSON.stringify({
+                    model: this.model,
+                    messages: [{ role: "user", content: prompt }],
+                    stream: false,
+                    options: { num_ctx: numCtx },
+                }),
+                signal: controller.signal,
+            });
+            if (!response.ok) {
+                const errorBody = await response.text();
+                throw new Error(`Ollama error (${response.status}): ${errorBody}`);
+            }
+            const data = (await response.json());
+            const content = data.message?.content;
+            if (!content) {
+                throw new Error("Ollama returned empty response");
+            }
+            return content;
+        }
+        catch (error) {
+            throw classifyError(error, this.name);
+        }
+        finally {
+            clearTimeout(timer);
+        }
+    }
+    /**
+     * OpenAI-compatible API — used by LM Studio and unknown sources.
+     * Parses LM Studio-specific context window errors into actionable messages.
+     */
+    async executeOpenAICompat(prompt, timeoutMs) {
+        const timeout = timeoutMs ?? this.defaultTimeout;
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), timeout);
+        try {
+            // Don't send max_tokens — let the server auto-determine based on remaining
+            // context window. Sending max_tokens risks prompt + max_tokens > context_length
+            // which LM Studio rejects outright.
+            const response = await fetch(`${this.endpoint}/v1/chat/completions`, {
+                method: "POST",
+                headers: { "Content-Type": "application/json" },
+                body: JSON.stringify({
+                    model: this.model,
+                    messages: [{ role: "user", content: prompt }],
+                }),
+                signal: controller.signal,
+            });
+            if (!response.ok) {
+                const errorBody = await response.text();
+                // Detect LM Studio context window overflow
+                if (this.source === "LM Studio" && this.isContextWindowError(errorBody)) {
+                    const estimatedTokens = Math.ceil(prompt.length / 4);
+                    throw new Error(`LM Studio context window too small (prompt needs ~${estimatedTokens} tokens).\n` +
+                        `  Fix: In LM Studio → Developer → Model Settings → increase Context Length.\n` +
+                        `  Also uncheck the Response Limit, or set it higher.\n` +
+                        `  Note: higher context = more VRAM. Find the sweet spot for your GPU.`);
+                }
+                throw new Error(`Local LLM error (${response.status}): ${errorBody}`);
+            }
+            const data = (await response.json());
+            const content = data.choices?.[0]?.message?.content;
+            if (!content) {
+                throw new Error("Local LLM returned empty response");
+            }
+            return content;
+        }
+        catch (error) {
+            throw classifyError(error, this.name);
+        }
+        finally {
+            clearTimeout(timer);
+        }
+    }
+    /**
+     * Detect Ollama model context window via /api/show.
+     * Looks for context_length in model_info (e.g. "llama.context_length": 32768).
+     */
+    async detectOllamaContext() {
+        try {
+            const controller = new AbortController();
+            const timer = setTimeout(() => controller.abort(), 5000);
+            const response = await fetch(`${this.endpoint}/api/show`, {
+                method: "POST",
+                headers: { "Content-Type": "application/json" },
+                body: JSON.stringify({ name: this.model }),
+                signal: controller.signal,
+            });
+            clearTimeout(timer);
+            if (!response.ok)
+                return undefined;
+            const data = (await response.json());
+            if (data.model_info) {
+                for (const [key, value] of Object.entries(data.model_info)) {
+                    if (key.endsWith(".context_length") && typeof value === "number") {
+                        return value;
+                    }
+                }
+            }
+            return undefined;
+        }
+        catch {
+            return undefined;
+        }
+    }
+    /**
+     * Detect LM Studio context window errors from response body.
+     */
+    isContextWindowError(errorBody) {
+        const lower = errorBody.toLowerCase();
+        return ((lower.includes("n_keep") && lower.includes("n_ctx")) ||
+            lower.includes("context length exceeded") ||
+            lower.includes("maximum context length") ||
+            lower.includes("too many tokens"));
+    }
+}
+//# sourceMappingURL=local-llm.js.map

package/dist/adapters/local-llm.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"local-llm.js","sourceRoot":"","sources":["../../src/adapters/local-llm.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,WAAW,CAAC;AACxC,OAAO,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAInD,MAAM,OAAO,eAAgB,SAAQ,WAAW;IACrC,IAAI,CAAS;IAEd,QAAQ,CAAS;IACjB,KAAK,CAAS;IACd,MAAM,CAAc;IACpB,cAAc,CAAS;IACvB,qBAAqB,CAAqB;IAElD,YAAY,QAAgB,EAAE,KAAa,EAAE,IAAY,EAAE,MAAoB,EAAE,YAAoB,OAAO;QAC1G,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,CAAC,uBAAuB;QACrE,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;QACjB,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,cAAc,GAAG,SAAS,CAAC;IAClC,CAAC;IAED,KAAK,CAAC,WAAW;QACf,IAAI,CAAC;YACH,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;YACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,IAAI,CAAC,CAAC;YACzD,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,YAAY,EAAE;gBACzD,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAC;YACH,YAAY,CAAC,KAAK,CAAC,CAAC;YACpB,OAAO,QAAQ,CAAC,EAAE,CAAC;QACrB,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,KAAK,CAAC;QACf,CAAC;IACH,CAAC;IAED;;;;OAIG;IACH,KAAK,CAAC,mBAAmB;QACvB,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC7B,IAAI,CAAC,qBAAqB,GAAG,MAAM,IAAI,CAAC,mBAAmB,EAAE,CAAC;QAChE,CAAC;QACD,uDAAuD;QACvD,OAAO,IAAI,CAAC,qBAAqB,CAAC;IACpC,CAAC;IAED;;;;;;OAMG;IACM,iBAAiB;QACxB,8DAA8D;QAC9D,MAAM,aAAa,GAAG,IAAI,CAAC,qBAAqB;eAC3C,CAAC,IAAI,CAAC,MAAM,KAAK,WAAW,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;QAEvD,IAAI,CAAC,aAAa;YAAE,OAAO,SAAS,CAAC;QAErC,MAAM,eAAe,GAAG,IAAI,CAAC;QAC7B,MAAM,eAAe,GAAG,IAAI,CAAC,CAAC,8CAA8C;QAC5E,MAAM,eAAe,GAAG,IAAI,CAAC,GAAG,CAAC,aAAa,GAAG,eAAe,GAAG,eAAe,EAAE,IAAI,CAAC,CAAC;QAC1F,sCAAsC;QACtC,OAAO,eAAe,GAAG,CAAC,CAAC;IAC7B,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,MAAc,EAAE,SAAkB;QAC9C,IAAI,CAAC;YACH,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;gBAC7B,OAAO,MAAM,IAAI,CAAC,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;YACrD,CAAC;YACD,OAAO,MAAM,IAAI,CAAC,mBAAmB,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAC3D,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,iGAAiG;YACjG,IAAI,KAAK,YAAY,KAAK,IAAI,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;gBACvE,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;gBAC9C,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;oBAC7B,OAAO,IAAI,CAAC,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;gBAC/C,CAAC;gBACD,OAAO,IAAI,CAAC,mBAAmB,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;YACrD,CAAC;YACD,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAED;;;OAGG;IACK,KAAK,CAAC,aAAa,CAAC,MAAc,EAAE,SAAkB;QAC5D,MAAM,OAAO,GAAG,SAAS,IAAI,IAAI,CAAC,cAAc,CAAC;QACjD,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,OAAO,CAAC,CAAC;QAE5D,mEAAmE;QACnE,MAAM,qBAAqB,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAC3D,MAAM,cAAc,GAAG,IAAI,CAAC;QAC5B,MAAM,YAAY,GAAG,GAAG,CAAC;QACzB,IAAI,MAAM,GAAG,qBAAqB,GAAG,cAAc,GAAG,YAAY,CAAC;QAEnE,2CAA2C;QAC3C,IAAI,IAAI,CAAC,qBAAqB,EAAE,CAAC;YAC/B,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,IAAI,CAAC,qBAAqB,CAAC,CAAC;QACxD,CAAC;QAED,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,WAAW,EAAE;gBACxD,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;gBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACnB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;oBAC7C,MAAM,EAAE,KAAK;oBACb,OAAO,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE;iBAC7B,CAAC;gBACF,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAC;YAEH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBACjB,MAAM,SAAS,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;gBACxC,MAAM,IAAI,KAAK,CAAC,iBAAiB,QAAQ,CAAC,MAAM,MAAM,SAAS,EAAE,CAAC,CAAC;YACrE,CAAC;YAED,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAElC,CAAC;YAEF,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC;YACtC,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;YACpD,CAAC;YAED,OAAO,OAAO,CAAC;QACjB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,CAAC,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QACxC,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACtB,CAAC;IACH,CAAC;IAED;;;OAGG;IACK,KAAK,CAAC,mBAAmB,CAAC,MAAc,EAAE,SAAkB;QAClE,MAAM,OAAO,GAAG,SAAS,IAAI,IAAI,CAAC,cAAc,CAAC;QACjD,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,OAAO,CAAC,CAAC;QAE5D,IAAI,CAAC;YACH,2EAA2E;YAC3E,gFAAgF;YAChF,oCAAoC;YACpC,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,sBAAsB,EAAE;gBACnE,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;gBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACnB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;iBAC9C,CAAC;gBACF,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAC;YAEH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBACjB,MAAM,SAAS,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;gBACxC,2CAA2C;gBAC3C,IAAI,IAAI,CAAC,MAAM,KAAK,WAAW,IAAI,IAAI,CAAC,oBAAoB,CAAC,SAAS,CAAC,EAAE,CAAC;oBACxE,MAAM,eAAe,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;oBACrD,MAAM,IAAI,KAAK,CACb,qDAAqD,eAAe,aAAa;wBACjF,+EAA+E;wBAC/E,wDAAwD;wBACxD,uEAAuE,CACxE,CAAC;gBACJ,CAAC;gBACD,MAAM,IAAI,KAAK,CAAC,oBAAoB,QAAQ,CAAC,MAAM,MAAM,SAAS,EAAE,CAAC,CAAC;YACxE,CAAC;YAED,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAElC,CAAC;YAEF,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC;YACpD,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,MAAM,IAAI,KAAK,CAAC,mCAAmC,CAAC,CAAC;YACvD,CAAC;YAED,OAAO,OAAO,CAAC;QACjB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,CAAC,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QACxC,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACtB,CAAC;IACH,CAAC;IAED;;;OAGG;IACK,KAAK,CAAC,mBAAmB;QAC/B,IAAI,CAAC;YACH,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;YACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,IAAI,CAAC,CAAC;YACzD,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,WAAW,EAAE;gBACxD,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;gBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC;gBAC1C,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAC;YACH,YAAY,CAAC,KAAK,CAAC,CAAC;YAEpB,IAAI,CAAC,QAAQ,CAAC,EAAE;gBAAE,OAAO,SAAS,CAAC;YAEnC,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAElC,CAAC;YAEF,IAAI,IAAI,CAAC,UAAU,EAAE,CAAC;gBACpB,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,EAAE,CAAC;oBAC3D,IAAI,GAAG,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;wBACjE,OAAO,KAAK,CAAC;oBACf,CAAC;gBACH,CAAC;YACH,CAAC;YAED,OAAO,SAAS,CAAC;QACnB,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,SAAS,CAAC;QACnB,CAAC;IACH,CAAC;IAED;;OAEG;IACK,oBAAoB,CAAC,SAAiB;QAC5C,MAAM,KAAK,GAAG,SAAS,CAAC,WAAW,EAAE,CAAC;QACtC,OAAO,CACL,CAAC,KAAK,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;YACrD,KAAK,CAAC,QAAQ,CAAC,yBAAyB,CAAC;YACzC,KAAK,CAAC,QAAQ,CAAC,wBAAwB,CAAC;YACxC,KAAK,CAAC,QAAQ,CAAC,iBAAiB,CAAC,CAClC,CAAC;IACJ,CAAC;CACF"}

package/dist/adapters/openai-api.js

@@ -34,7 +34,7 @@ export class OpenAIApiAdapter extends BaseAdapter {
                 body: JSON.stringify({
                     model: this.model,
                     messages: [{ role: "user", content: prompt }],
-                    max_tokens: 16384,
+                    max_completion_tokens: 16384,
                 }),
                 signal: controller.signal,
             });

package/dist/adapters/openai-api.js.map

@@ -1 +1 @@
-{"version":3,"file":"openai-api.js","sourceRoot":"","sources":["../../src/adapters/openai-api.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,WAAW,CAAC;AACxC,OAAO,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AACnD,OAAO,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAE1C,MAAM,OAAO,gBAAiB,SAAQ,WAAW;IACtC,IAAI,GAAG,KAAK,CAAC;IAEd,KAAK,CAAS;IACd,MAAM,CAAS;IACf,cAAc,CAAS;IAE/B,YAAY,QAAgB,SAAS,EAAE,SAAiB,gBAAgB,EAAE,YAAoB,OAAO;QACnG,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,cAAc,GAAG,SAAS,CAAC;IAClC,CAAC;IAED,KAAK,CAAC,WAAW;QACf,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACtC,OAAO,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC;IACnD,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,MAAc,EAAE,SAAkB;QAC9C,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACzC,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,MAAM,aAAa,CACjB,IAAI,KAAK,CAAC,+BAA+B,IAAI,CAAC,MAAM,4BAA4B,CAAC,EACjF,IAAI,CAAC,IAAI,CACV,CAAC;QACJ,CAAC;QAED,MAAM,OAAO,GAAG,SAAS,IAAI,IAAI,CAAC,cAAc,CAAC;QACjD,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,OAAO,CAAC,CAAC;QAE5D,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,4CAA4C,EAAE;gBACzE,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE;oBACP,cAAc,EAAE,kBAAkB;oBAClC,aAAa,EAAE,UAAU,MAAM,EAAE;iBAClC;gBACD,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACnB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;oBAC7C,UAAU,EAAE,KAAK;iBAClB,CAAC;gBACF,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAC;YAEH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBACjB,MAAM,SAAS,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;gBACxC,MAAM,IAAI,KAAK,CAAC,qBAAqB,QAAQ,CAAC,MAAM,MAAM,SAAS,EAAE,CAAC,CAAC;YACzE,CAAC;YAED,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAElC,CAAC;YAEF,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC;YACpD,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;YACxD,CAAC;YAED,OAAO,OAAO,CAAC;QACjB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,CAAC,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QACxC,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACtB,CAAC;IACH,CAAC;CACF"}
+{"version":3,"file":"openai-api.js","sourceRoot":"","sources":["../../src/adapters/openai-api.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,WAAW,CAAC;AACxC,OAAO,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AACnD,OAAO,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAE1C,MAAM,OAAO,gBAAiB,SAAQ,WAAW;IACtC,IAAI,GAAG,KAAK,CAAC;IAEd,KAAK,CAAS;IACd,MAAM,CAAS;IACf,cAAc,CAAS;IAE/B,YAAY,QAAgB,SAAS,EAAE,SAAiB,gBAAgB,EAAE,YAAoB,OAAO;QACnG,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,cAAc,GAAG,SAAS,CAAC;IAClC,CAAC;IAED,KAAK,CAAC,WAAW;QACf,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACtC,OAAO,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC;IACnD,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,MAAc,EAAE,SAAkB;QAC9C,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACzC,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,MAAM,aAAa,CACjB,IAAI,KAAK,CAAC,+BAA+B,IAAI,CAAC,MAAM,4BAA4B,CAAC,EACjF,IAAI,CAAC,IAAI,CACV,CAAC;QACJ,CAAC;QAED,MAAM,OAAO,GAAG,SAAS,IAAI,IAAI,CAAC,cAAc,CAAC;QACjD,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,OAAO,CAAC,CAAC;QAE5D,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,4CAA4C,EAAE;gBACzE,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE;oBACP,cAAc,EAAE,kBAAkB;oBAClC,aAAa,EAAE,UAAU,MAAM,EAAE;iBAClC;gBACD,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACnB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;oBAC7C,qBAAqB,EAAE,KAAK;iBAC7B,CAAC;gBACF,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAC;YAEH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBACjB,MAAM,SAAS,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;gBACxC,MAAM,IAAI,KAAK,CAAC,qBAAqB,QAAQ,CAAC,MAAM,MAAM,SAAS,EAAE,CAAC,CAAC;YACzE,CAAC;YAED,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAElC,CAAC;YAEF,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC;YACpD,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;YACxD,CAAC;YAED,OAAO,OAAO,CAAC;QACjB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,aAAa,CAAC,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QACxC,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACtB,CAAC;IACH,CAAC;CACF"}

package/dist/commands/init.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"init.d.ts","sourceRoot":"","sources":["../../src/commands/init.ts"],"names":[],"mappings":"AAiLA;;GAEG;AACH,wBAAsB,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAwMhE"}
+{"version":3,"file":"init.d.ts","sourceRoot":"","sources":["../../src/commands/init.ts"],"names":[],"mappings":"AA4NA;;GAEG;AACH,wBAAsB,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAsNhE"}

package/dist/commands/init.js

@@ -6,6 +6,7 @@ import { execa } from "execa";
 import chalk from "chalk";
 import ora from "ora";
 import { saveKey, getKey, getKeysPath } from "../utils/keys.js";
+import { detectLocalModels } from "../utils/local-detect.js";
 const rl = () => createInterface({ input: process.stdin, output: process.stdout });
 /**
  * Ask a yes/no question. Returns true for yes.
@@ -105,14 +106,61 @@ const DEFAULT_CAPABILITIES = {
 /**
  * Generate a config.json based on detected tools and user choices.
  */
-function generateConfig(projectName, language, enabledKnights) {
-    const knights = enabledKnights.map((k, i) => ({
+function generateConfig(projectName, language, enabledKnights, localKnights = []) {
+    let priority = 1;
+    const knights = enabledKnights.map((k) => ({
         name: k.name,
         adapter: k.adapter,
         capabilities: DEFAULT_CAPABILITIES[k.name] || ["general"],
-        priority: i + 1,
+        priority: priority++,
         ...(k.fallback ? { fallback: k.fallback } : {}),
     }));
+    // Append local knights
+    for (const lk of localKnights) {
+        knights.push({
+            name: lk.name,
+            adapter: lk.adapter,
+            capabilities: DEFAULT_CAPABILITIES[lk.name] || ["code", "logic"],
+            priority: priority++,
+        });
+    }
+    // Build adapter_config — start with cloud adapters
+    const adapter_config = {
+        "claude-cli": {
+            command: "claude",
+            args: ["-p", "{prompt}", "--print"],
+        },
+        "claude-api": {
+            model: "claude-sonnet-4-6",
+            env_key: "ANTHROPIC_API_KEY",
+        },
+        "gemini-cli": {
+            command: "gemini",
+            args: ["-p", "{prompt}"],
+        },
+        "gemini-api": {
+            model: "gemini-2.5-flash",
+            env_key: "GEMINI_API_KEY",
+        },
+        "openai-cli": {
+            command: "codex",
+            args: ["exec", "{prompt}"],
+        },
+        "openai-api": {
+            model: "gpt-5.2",
+            env_key: "OPENAI_API_KEY",
+        },
+    };
+    // Add local adapter configs
+    for (const lk of localKnights) {
+        const localCfg = {
+            endpoint: lk.endpoint,
+            model: lk.model,
+            name: lk.name,
+            source: lk.source,
+        };
+        adapter_config[lk.adapter] = localCfg;
+    }
     return {
         version: "1.0",
         project: projectName,
@@ -127,32 +175,7 @@ function generateConfig(projectName, language, enabledKnights) {
             ignore: [".git", "node_modules", "dist", "build", ".next"],
         },
         chronicle: ".roundtable/chronicle.md",
-        adapter_config: {
-            "claude-cli": {
-                command: "claude",
-                args: ["-p", "{prompt}", "--print"],
-            },
-            "claude-api": {
-                model: "claude-sonnet-4-6",
-                env_key: "ANTHROPIC_API_KEY",
-            },
-            "gemini-cli": {
-                command: "gemini",
-                args: ["-p", "{prompt}"],
-            },
-            "gemini-api": {
-                model: "gemini-2.5-flash",
-                env_key: "GEMINI_API_KEY",
-            },
-            "openai-cli": {
-                command: "codex",
-                args: ["exec", "{prompt}"],
-            },
-            "openai-api": {
-                model: "gpt-5.2",
-                env_key: "OPENAI_API_KEY",
-            },
-        },
+        adapter_config,
     };
 }
 /**
@@ -178,10 +201,10 @@ export async function initCommand(version) {
     const projectName = await askText("  Project name?", dirName);
     // 2. Language
     const language = await askText("  Discussion language?", "nl");
-    // 3. Detect tools
+    // 3. Detect all knights (CLI tools + local LLM servers)
     console.log("");
     const detectSpinner = ora("  Scouting for available knights...").start();
-    const tools = await detectTools();
+    const [tools, localModels] = await Promise.all([detectTools(), detectLocalModels()]);
     detectSpinner.succeed("  Scouting complete");
     for (const tool of tools) {
         const icon = tool.available ? chalk.green("  +") : chalk.red("  -");
@@ -190,6 +213,9 @@ export async function initCommand(version) {
             : chalk.dim("not found");
         console.log(`${icon} ${chalk.bold(tool.name)} ${chalk.dim(`(${tool.command})`)} — ${status}`);
     }
+    for (const model of localModels) {
+        console.log(`${chalk.green("  +")} ${chalk.bold(model.name)} ${chalk.dim(`(${model.source})`)} — ${chalk.green("local")}`);
+    }
     // 4. Let user choose which knights to enable
     console.log("");
     const enabledKnights = [];
@@ -220,21 +246,13 @@ export async function initCommand(version) {
                     adapter: tool.adapter,
                     fallback: FALLBACKS[tool.adapter],
                 });
-                // Offer to set up fallback API key
+                // Fallback API key — only ask once, remember forever
                 const fallbackAdapter = FALLBACKS[tool.adapter];
                 const fallbackEnvKey = fallbackAdapter ? API_ENV_KEYS[fallbackAdapter] : undefined;
                 if (fallbackEnvKey) {
                     const existingFallbackKey = await getKey(fallbackEnvKey);
                     if (existingFallbackKey) {
-                        console.log(chalk.dim(`  ✓ ${tool.name} fallback API key already set`));
-                        const update = await confirm(`  Replace existing ${tool.name} fallback API key?`, false);
-                        if (update) {
-                            const key = await askSecret(`  Enter your new ${tool.name} API key:`);
-                            if (key) {
-                                await saveKey(fallbackEnvKey, key);
-                                console.log(chalk.green(`  ✓ ${tool.name} fallback API key updated`));
-                            }
-                        }
+                        console.log(chalk.green(`  ✓ ${tool.name} fallback API key already set`));
                     }
                     else {
                         const wantFallback = await confirm(`  Set up a fallback API key for ${tool.name}? (used if CLI fails)`, false);
@@ -256,25 +274,15 @@ export async function initCommand(version) {
                 const apiAdapter = API_ADAPTERS[tool.adapter] || tool.adapter;
                 const envKey = API_ENV_KEYS[apiAdapter];
                 if (envKey) {
-                    // Check if key already exists (env var or keystore)
                     const existingKey = await getKey(envKey);
                     if (existingKey) {
-                        console.log(chalk.green(`  ✓ ${tool.name} API key found`));
-                        const update = await confirm(`  Replace existing ${tool.name} API key?`, false);
-                        if (update) {
-                            const key = await askSecret(`  Enter your new ${tool.name} API key:`);
-                            if (key) {
-                                await saveKey(envKey, key);
-                                console.log(chalk.green(`  ✓ ${tool.name} API key updated in ${chalk.dim(getKeysPath())}`));
-                            }
-                        }
+                        console.log(chalk.green(`  ✓ ${tool.name} API key already set`));
                     }
                     else {
-                        // Ask for the API key
                         const key = await askSecret(`  Enter your ${tool.name} API key:`);
                         if (key) {
                             await saveKey(envKey, key);
-                            console.log(chalk.green(`  ✓ ${tool.name} API key saved securely to ${chalk.dim(getKeysPath())}`));
+                            console.log(chalk.green(`  ✓ ${tool.name} API key saved to ${chalk.dim(getKeysPath())}`));
                         }
                         else {
                             apiKeyReminders.push(`  ${envKey}  # ${tool.name}`);
@@ -288,13 +296,38 @@ export async function initCommand(version) {
             }
         }
     }
-    if (enabledKnights.length === 0) {
+    // 4b. Seat local LLM models (already detected in step 3)
+    const enabledLocalKnights = [];
+    for (const model of localModels) {
+        const use = await confirm(`  Seat ${model.name} at the table?`, true);
+        if (use) {
+            const slug = model.name.toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/(^-|-$)/g, "");
+            enabledLocalKnights.push({
+                name: model.name,
+                adapter: `local-llm-${slug}`,
+                endpoint: model.endpoint,
+                model: model.modelId,
+                source: model.source,
+            });
+        }
+    }
+    // Show LM Studio setup tip for each seated LM Studio model
+    const lmStudioKnights = enabledLocalKnights.filter((k) => k.source === "LM Studio");
+    if (lmStudioKnights.length > 0) {
+        console.log(chalk.yellow("\n  ⚠  LM Studio setup required:"));
+        console.log(chalk.white("     The default Context Length (4096) is too small for roundtable prompts."));
+        console.log(chalk.white("     → Context Length: increase to at least 16384, or as high as your model/GPU allows"));
+        console.log(chalk.white("     → Response Limit: uncheck the limit, or set to 4096+"));
+        console.log(chalk.dim("     Find these in: Developer tab → Model Settings"));
+        console.log(chalk.dim("     Note: higher context = more VRAM + slower. Find the sweet spot for your setup."));
+    }
+    if (enabledKnights.length === 0 && enabledLocalKnights.length === 0) {
         console.log(chalk.red("\n  A roundtable with no knights is just a table."));
         console.log(chalk.dim("  Re-run `roundtable init` and enable at least one knight."));
         return;
     }
     // 5. Generate config
-    const config = generateConfig(projectName, language, enabledKnights);
+    const config = generateConfig(projectName, language, enabledKnights, enabledLocalKnights);
     // 6. Create folder structure
     console.log("");
     const structureSpinner = ora("  Forging the roundtable...").start();