lynkr 9.3.3 → 9.4.1

package/.impeccable/live/config.json

@@ -0,0 +1,8 @@
+{
+  "files": ["docs/**/*.html"],
+  "insertBefore": "</body>",
+  "commentSyntax": "html",
+  "cspChecked": true,
+  "framework": "multi-page-static",
+  "devServer": null
+}

package/README.md

@@ -1,22 +1,27 @@
 # Lynkr
 
-### Run Claude Code, Cursor, and Codex on any model. One proxy, every provider.
+### The AI coding proxy that compresses tokens before they hit the model.
+
+**87.6% fewer tokens on JSON tool results. 53% fewer tokens on tool-heavy requests. 171ms semantic cache hits. Zero code changes.**
 
 [![npm version](https://img.shields.io/npm/v/lynkr.svg)](https://www.npmjs.com/package/lynkr)
 [![Tests](https://img.shields.io/badge/tests-699%20passing-brightgreen)](https://github.com/Fast-Editor/Lynkr)
 [![License: Apache 2.0](https://img.shields.io/badge/license-Apache%202.0-blue.svg)](LICENSE)
 [![Node.js](https://img.shields.io/badge/node-20%2B-green)](https://nodejs.org)
-[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/vishalveerareddy123/Lynkr)
+[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/Fast-Editor/Lynkr)
 
 <table>
 <tr>
-<td align="center"><strong>12+</strong><br/>LLM Providers</td>
-<td align="center"><strong>60-80%</strong><br/>Cost Reduction</td>
-<td align="center"><strong>699</strong><br/>Tests Passing</td>
+<td align="center"><strong>87.6%</strong><br/>JSON Compression</td>
+<td align="center"><strong>53%</strong><br/>Tool Token Reduction</td>
+<td align="center"><strong>171ms</strong><br/>Semantic Cache Hits</td>
+<td align="center"><strong>13+</strong><br/>LLM Providers</td>
 <td align="center"><strong>0</strong><br/>Code Changes Required</td>
 </tr>
 </table>
 
+> Numbers from a live benchmark against LiteLLM on identical workloads. [See full report →](BENCHMARK_REPORT.md)
+
 ---
 
 ## Quick Start (2 Minutes)
@@ -27,201 +32,63 @@
 npm install -g lynkr
 ```
 
-### 2. Configure for Ollama (Free & Local)
+### 2. Configure Lynkr
 
-**Step 2a: Install and start Ollama**
+First run creates a `.env` file. Edit it with your provider settings.
 
-Download from https://ollama.com and install, then:
+**Option A: Free & Local (Ollama) - Recommended for Testing**
 
 ```bash
+# Install Ollama first: https://ollama.com
 ollama pull qwen2.5-coder:latest
-ollama serve
-```
-
-**Step 2b: Set environment variables**
-
-**Windows (Command Prompt):**
-```cmd
-set MODEL_PROVIDER=ollama
-set FALLBACK_ENABLED=false
-set OLLAMA_MODEL=qwen2.5-coder:latest
-set OLLAMA_ENDPOINT=http://localhost:11434
-set PORT=8081
-```
-
-**Windows (PowerShell):**
-```powershell
-$env:MODEL_PROVIDER="ollama"
-$env:FALLBACK_ENABLED="false"
-$env:OLLAMA_MODEL="qwen2.5-coder:latest"
-$env:OLLAMA_ENDPOINT="http://localhost:11434"
-$env:PORT="8081"
 ```
 
-**Linux/macOS:**
+Create/edit `.env` in your project directory:
 ```bash
-export MODEL_PROVIDER=ollama
-export FALLBACK_ENABLED=false
-export OLLAMA_MODEL=qwen2.5-coder:latest
-export OLLAMA_ENDPOINT=http://localhost:11434
-export PORT=8081
-```
-
-**Alternative: Use a .env file (recommended for permanent config)**
-
-Create a file named `.env` in your home directory or project folder:
-
-```bash
-# Required: Provider Configuration
+# Provider
 MODEL_PROVIDER=ollama
 FALLBACK_ENABLED=false
 
-# Required: Ollama Settings
-OLLAMA_MODEL=qwen2.5-coder:latest
+# Ollama Configuration
 OLLAMA_ENDPOINT=http://localhost:11434
+OLLAMA_MODEL=qwen2.5-coder:latest
 
-# Required: Server
+# Server
 PORT=8081
-HOST=0.0.0.0
-```
-
-### 3. Start Lynkr
 
-```bash
-lynkr start
-```
-
-You should see:
-```
-[Ollama] Server ready, model "qwen2.5-coder:latest" available
-Claude→Databricks proxy listening on http://localhost:8081
-```
-
-### 4. Connect Claude Code (or your AI tool)
-
-**Windows (Command Prompt):**
-```cmd
-set ANTHROPIC_BASE_URL=http://localhost:8081
-set ANTHROPIC_API_KEY=dummy
-claude "write hello world in python"
-```
+# Optional: Limits (remove for unlimited)
+POLICY_MAX_STEPS=50
+POLICY_MAX_TOOL_CALLS=100
 
-**Linux/macOS:**
-```bash
-export ANTHROPIC_BASE_URL=http://localhost:8081
-export ANTHROPIC_API_KEY=dummy
-claude "write hello world in python"
+# Disable overly strict command filtering
+POLICY_SAFE_COMMANDS_ENABLED=false
 ```
 
-✅ **Done!** Claude Code now uses your local Ollama model.
-
----
-
-## Alternative Setup: Cloud Provider (OpenRouter)
-
-Get API key from https://openrouter.ai, then:
-
-**Windows (Command Prompt):**
-```cmd
-set MODEL_PROVIDER=openrouter
-set OPENROUTER_API_KEY=sk-or-v1-your-key-here
-set FALLBACK_ENABLED=false
-set PORT=8081
-lynkr start
-```
+**Option B: Cloud (OpenRouter) - Recommended for Production**
 
-**Linux/macOS:**
 ```bash
-export MODEL_PROVIDER=openrouter
-export OPENROUTER_API_KEY=sk-or-v1-your-key-here
-export FALLBACK_ENABLED=false
-export PORT=8081
-lynkr start
+# Get API key from https://openrouter.ai
 ```
 
-**Or use .env file:**
+Create/edit `.env`:
 ```bash
+# Provider
 MODEL_PROVIDER=openrouter
 OPENROUTER_API_KEY=sk-or-v1-your-key-here
 FALLBACK_ENABLED=false
-PORT=8081
-```
-
----
-
-## Common Startup Errors (READ THIS IF IT DOESN'T WORK!)
-
-### Error: `unable to determine transport target for "pino-pretty"`
-
-**Problem:** This error happens on Windows or when NODE_ENV is not set.
-
-**Solution:** Set NODE_ENV before starting:
-
-**Windows (Command Prompt):**
-```cmd
-set NODE_ENV=production
-lynkr start
-```
-
-**Windows (PowerShell):**
-```powershell
-$env:NODE_ENV="production"
-lynkr start
-```
-
-**Linux/macOS:**
-```bash
-export NODE_ENV=production
-lynkr start
-```
-
-Or add to your `.env` file:
-```bash
-NODE_ENV=production
-```
-
-### Warning: `Missing tier configuration: TIER_SIMPLE, TIER_MEDIUM...`
-
-**This is just a warning - you can ignore it.** Tier routing is optional for advanced use.
-
-To remove the warning, add these to your environment or `.env`:
-
-```bash
-TIER_SIMPLE=ollama:qwen2.5-coder:latest
-TIER_MEDIUM=ollama:qwen2.5-coder:latest
-TIER_COMPLEX=ollama:qwen2.5-coder:latest
-TIER_REASONING=ollama:qwen2.5-coder:latest
-```
-
-(Use the same model for all tiers if you only have one model)
-
-### Warning: `FALLBACK_PROVIDER='databricks' is enabled but missing credentials`
 
-**Solution:** Disable fallback if you're only using one provider:
+# Server
+PORT=8081
 
-```bash
-set FALLBACK_ENABLED=false  # Windows
-export FALLBACK_ENABLED=false  # Linux/macOS
-```
+# Optional: Limits (remove for unlimited)
+POLICY_MAX_STEPS=50
+POLICY_MAX_TOOL_CALLS=100
 
-Or add to `.env`:
-```bash
-FALLBACK_ENABLED=false
+# Optional: Enable caching
+PROMPT_CACHE_ENABLED=true
+SEMANTIC_CACHE_ENABLED=true
 ```
 
-### Error: `Ollama connection refused`
-
-**Problem:** Ollama is not running.
-
-**Solution:**
-1. Check if Ollama is running: `ollama list`
-2. Start Ollama: `ollama serve` (or restart the Ollama app)
-3. Verify model is pulled: `ollama pull qwen2.5-coder:latest`
-
----
-
-## Additional Provider Examples
-
 **Option C: Enterprise (AWS Bedrock)**
 
 Create/edit `.env`:
@@ -259,6 +126,7 @@ POLICY_MAX_TOOL_CALLS=100
 ```
 
 Then start Lynkr:
+
 ```bash
 lynkr start
 ```
@@ -266,6 +134,15 @@ lynkr start
 ### 3. Connect Your Tool
 
 **Claude Code**
+
+**Windows (Command Prompt):**
+```cmd
+set ANTHROPIC_BASE_URL=http://localhost:8081
+set ANTHROPIC_API_KEY=dummy
+claude "write a hello world in python"
+```
+
+**Linux/macOS:**
 ```bash
 export ANTHROPIC_BASE_URL=http://localhost:8081
 export ANTHROPIC_API_KEY=dummy
@@ -292,23 +169,87 @@ wire_api = "responses"
 
 ---
 
+## Common Startup Errors
+
+### Error: `unable to determine transport target for "pino-pretty"`
+
+**Problem:** You're running an older version (< 9.3.0).
+
+**Solution:** Update to the latest version:
+```bash
+npm install -g lynkr@latest
+```
+
+If you must use an older version, set `NODE_ENV=production` before starting.
+
+### Warning: `Missing tier configuration: TIER_SIMPLE, TIER_MEDIUM...`
+
+**This is just a warning - you can ignore it.** Tier routing is optional.
+
+To remove the warning, add to `.env`:
+```bash
+TIER_SIMPLE=ollama:qwen2.5-coder:latest
+TIER_MEDIUM=ollama:qwen2.5-coder:latest
+TIER_COMPLEX=ollama:qwen2.5-coder:latest
+TIER_REASONING=ollama:qwen2.5-coder:latest
+```
+
+### Warning: `FALLBACK_PROVIDER='databricks' is enabled but missing credentials`
+
+**Solution:** Add to `.env`:
+```bash
+FALLBACK_ENABLED=false
+```
+
+### Error: `connect ECONNREFUSED ::1:11434` (Ollama)
+
+**Problem:** Ollama is not running.
+
+**Solution:**
+```bash
+ollama serve
+```
+
+Keep this terminal open, and start Lynkr in a new terminal.
+
+### Error: `Connection refused` or `404 Not Found`
+
+**Problem:** Lynkr is not running or wrong port.
+
+**Solution:** Check Lynkr is running on the correct port:
+```bash
+curl http://localhost:8081/
+```
+
+Should return: `{"service":"Lynkr","version":"9.x.x","status":"running"}`
+
+---
+
 ## Why Lynkr?
 
-AI coding tools lock you into one provider. Lynkr breaks that lock.
+AI coding tools lock you into one provider and send every token raw. Lynkr breaks both locks.
 
 ```
 Claude Code / Cursor / Codex / Cline / Continue
                     ↓
                   Lynkr
+          ┌─────────────────────┐
+          │  Strip unused tools  │  ← 53% fewer tokens on tool calls
+          │  Compress JSON blobs │  ← 87.6% on large tool results
+          │  Semantic cache      │  ← 171ms hits, 0 tokens billed
+          │  Route by complexity │  ← cheap model for simple, cloud for hard
+          └─────────────────────┘
                     ↓
-    Ollama | Bedrock | Azure | OpenRouter | OpenAI
+    Ollama | Bedrock | Azure | Moonshot | OpenRouter | OpenAI
 ```
 
 **What you get:**
-- ✅ Use **free local models** (Ollama, llama.cpp) with Claude Code
+- ✅ **53% fewer tokens** on tool-heavy requests (Claude Code, Cursor sessions)
+- ✅ **87.6% compression** on large JSON tool results (grep, file reads, test output)
+- ✅ **Semantic cache** serves repeated queries in 171ms with 0 tokens billed
+- ✅ **Automatic tier routing** — simple questions go to cheap models, complex ones escalate
 - ✅ Route through **your company's infrastructure** (Databricks, Azure, Bedrock)
-- ✅ Cut costs **60-80%** with smart token optimization
-- ✅ **Zero code changes** - just change one environment variable
+- ✅ **Zero code changes** — just change one environment variable
 
 ---
 
@@ -670,7 +611,7 @@ curl -fsSL https://raw.githubusercontent.com/Fast-Editor/Lynkr/main/install.sh |
 
 **Homebrew**
 ```bash
-brew tap vishalveerareddy123/lynkr
+brew tap fast-editor/lynkr
 brew install lynkr
 ```
 
@@ -709,6 +650,42 @@ npm start
 
 ---
 
+## Benchmark Results
+
+Measured on real agentic coding workloads (Claude Code / Cursor sessions) with Ollama, Moonshot, and Azure OpenAI backends. Run with `node benchmark-tier-routing.js`.
+
+### Token compression
+
+| Scenario | Tokens without Lynkr | Tokens with Lynkr | Reduction |
+|---|---|---|---|
+| 14-tool request (read task) | 1,042 | **547** | **47%** |
+| 14-tool request (write task) | 1,043 | **412** | **60%** |
+| Large JSON grep result (60 items) | 3,458 | **427** | **87.6%** |
+
+Lynkr strips irrelevant tool schemas before forwarding (smart tool selection) and binary-compresses large JSON tool results (TOON) — both happen in-process with no added latency.
+
+### Semantic cache
+
+| | Tokens billed | Response time |
+|---|---|---|
+| First call (cold) | 2,857 | 1,891ms |
+| **Second call — paraphrased, cache hit** | **0** | **171ms** |
+
+Near-identical prompts return cached responses in 171ms. Zero tokens billed on a cache hit.
+
+### Tier routing
+
+| Request | Routed to |
+|---|---|
+| "What does git stash do?" | SIMPLE → local model (free) |
+| JWT vs cookies security analysis | COMPLEX → cloud model (correct) |
+
+Lynkr scores each request on 15 dimensions (token count, code complexity, reasoning markers, risk signals, agentic patterns) and routes automatically. No caller changes needed.
+
+→ [Full benchmark report with methodology](BENCHMARK_REPORT.md)
+
+---
+
 ## Cost Comparison
 
 | Scenario | Direct Anthropic | Lynkr + Ollama | Lynkr + OpenRouter |
@@ -716,7 +693,7 @@ npm start
 | Daily coding (8h) | $10-30/day | **$0 (free)** | $2-8/day |
 | Monthly (heavy use) | $300-900 | **$0** | $60-240 |
 
-With tier routing + token optimization: **additional 60-80% savings** on cloud providers.
+With tier routing + token optimization: **additional 50-87% savings** on cloud providers depending on workload.
 
 ---
 
@@ -727,13 +704,17 @@ With tier routing + token optimization: **additional 60-80% savings** on cloud p
 | **Setup** | `npm install -g lynkr` | Python + Docker + Postgres | Account signup | Docker stack |
 | **Claude Code native** | ✅ Drop-in | ⚠️ Requires config | ❌ | ⚠️ Partial |
 | **Cursor native** | ✅ Drop-in | ⚠️ Partial | ❌ | ⚠️ Partial |
-| **Local models** | Ollama, llama.cpp, LM Studio, MLX | Ollama only | ❌ | ❌ |
-| **Tier routing** | Auto complexity-based | ❌ Manual | Cost-based only | ❌ Manual |
-| **Token optimization** | 60-80% built-in | ❌ | ❌ | Cache only |
+| **Local models** | Ollama, llama.cpp, LM Studio | Ollama only | ❌ | ❌ |
+| **Automatic tier routing** | ✅ 15-dimension scorer | ⚠️ Cost-only | ❌ | ❌ Manual metadata |
+| **TOON JSON compression** | ✅ up to 87.6% | ❌ | ❌ | ❌ |
+| **Smart tool selection** | ✅ up to 60% token reduction | ❌ | ❌ | ❌ |
+| **Semantic cache** | ✅ 171ms hits, 0 tokens | ❌ | ❌ | ✅ Prompt cache only |
+| **Long-term memory** | ✅ SQLite, per-session | ❌ | ❌ | ❌ |
+| **MCP integration** | ✅ + Code Mode (96% reduction) | ❌ | ❌ | ❌ |
 | **Self-hosted** | ✅ Node.js only | ✅ Python stack | ❌ SaaS | ✅ Docker |
 | **Dependencies** | Node.js 20+ | Python, Prisma, PostgreSQL | None | Docker, Python |
 
-**Lynkr's edge:** Purpose-built for AI coding tools. Zero-config for Claude Code, Cursor, and Codex. Installs in one command, runs anywhere Node.js runs.
+**Lynkr's edge:** Purpose-built for AI coding tools. Compresses tokens before they reach the model — not just after. Zero-config for Claude Code, Cursor, and Codex. Installs in one command.
 
 ---
 
@@ -742,7 +723,7 @@ With tier routing + token optimization: **additional 60-80% savings** on cloud p
 - [GitHub Discussions](https://github.com/Fast-Editor/Lynkr/discussions) — Ask questions
 - [Report Issues](https://github.com/Fast-Editor/Lynkr/issues) — Bug reports
 - [NPM Package](https://www.npmjs.com/package/lynkr) — Official releases
-- [DeepWiki](https://deepwiki.com/vishalveerareddy123/Lynkr) — AI-powered docs
+- [DeepWiki](https://deepwiki.com/Fast-Editor/Lynkr) — AI-powered docs
 
 ---
 

package/benchmark-configs/litellm_config.yaml

@@ -0,0 +1,86 @@
+# ─── LiteLLM Benchmark Config ─────────────────────────────────────────────────
+# Multi-provider tier routing via LiteLLM Complexity Router.
+#
+# Start: litellm --config benchmark-configs/litellm_config.yaml --port 8082
+#
+# Required env vars:
+#   AZURE_OPENAI_API_KEY
+#   AZURE_OPENAI_ENDPOINT   (https://YOUR-RESOURCE.openai.azure.com)
+#   MOONSHOT_API_KEY
+#   (Ollama needs no key — running locally on :11434)
+#
+# Tier mapping (matches Lynkr benchmark config):
+#   SIMPLE    → ollama:minimax-m2.5:cloud
+#   MEDIUM    → ollama:minimax-m2.5:cloud
+#   COMPLEX   → moonshot:moonshot-v1-auto
+#   REASONING → azure-openai:gpt-5.2-chat
+
+model_list:
+
+  # ── SIMPLE + MEDIUM → Ollama minimax-m2.5:cloud ───────────────────────────
+  # Note: the model tag is "minimax-m2.5:cloud" — the colon is part of the
+  # Ollama model name, NOT a provider separator here.
+  - model_name: smart-router
+    litellm_params:
+      model: "ollama/minimax-m2.5:cloud"
+      api_base: http://localhost:11434
+  - model_name: smart-router
+    litellm_params:
+      model: "ollama/minimax-m2.5:cloud"
+      api_base: http://localhost:11434
+
+  # ── COMPLEX → Moonshot moonshot-v1-auto (matches Lynkr TIER_COMPLEX) ────────
+  - model_name: smart-router
+    litellm_params:
+      model: openai/moonshot-v1-auto
+      api_base: https://api.moonshot.ai/v1
+      api_key: os.environ/MOONSHOT_API_KEY
+
+  # ── REASONING → Azure OpenAI gpt-5.2-chat ─────────────────────────────────
+  - model_name: smart-router
+    litellm_params:
+      model: azure/gpt-5.2-chat
+      api_base: os.environ/AZURE_OPENAI_ENDPOINT
+      api_key: os.environ/AZURE_OPENAI_API_KEY
+      api_version: "2024-12-01-preview"
+
+  # ── Direct aliases (for targeted calls outside the benchmark) ─────────────
+  - model_name: ollama-minimax
+    litellm_params:
+      model: "ollama/minimax-m2.5:cloud"
+      api_base: http://localhost:11434
+
+  - model_name: moonshot-kimi-k2
+    litellm_params:
+      model: openai/moonshot-v1-auto
+      api_base: https://api.moonshot.ai/v1
+      api_key: os.environ/MOONSHOT_API_KEY
+
+  - model_name: azure-gpt5
+    litellm_params:
+      model: azure/gpt-5.2-chat
+      api_base: os.environ/AZURE_OPENAI_ENDPOINT
+      api_key: os.environ/AZURE_OPENAI_API_KEY
+      api_version: "2024-12-01-preview"
+
+router_settings:
+  routing_strategy: cost-based-routing
+  # Fallback: if smart-router fails on one deployment, try the next
+  fallbacks:
+    - smart-router:
+        - ollama-minimax
+        - moonshot-kimi-k2
+        - azure-gpt5
+  num_retries: 2
+  timeout: 90
+
+litellm_settings:
+  drop_params: true
+  use_responses_api: false
+  return_response_headers: true
+  success_callback: []
+  failure_callback: []
+
+general_settings:
+  master_key: sk-1234        # change this
+  port: 8082

package/benchmark-configs/lynkr.env

@@ -0,0 +1,48 @@
+# ─── Lynkr Benchmark Config ───────────────────────────────────────────────────
+# Multi-provider tier routing: Ollama → Moonshot → Azure OpenAI
+# Copy to .env and fill in your credentials.
+
+PORT=8081
+
+# ── Ollama (local, free) ───────────────────────────────────────────────────────
+OLLAMA_ENDPOINT=http://localhost:11434
+OLLAMA_MODEL=qwen2.5-coder:7b
+OLLAMA_TIMEOUT_MS=120000
+OLLAMA_EMBEDDINGS_MODEL=nomic-embed-text
+OLLAMA_EMBEDDINGS_ENDPOINT=http://localhost:11434/api/embeddings
+
+# ── Azure OpenAI ───────────────────────────────────────────────────────────────
+AZURE_OPENAI_ENDPOINT=https://YOUR-RESOURCE.openai.azure.com
+AZURE_OPENAI_API_KEY=your-azure-openai-key
+AZURE_OPENAI_DEPLOYMENT=gpt-4o
+AZURE_OPENAI_API_VERSION=2024-08-01-preview
+
+# ── Moonshot (Kimi) ────────────────────────────────────────────────────────────
+MOONSHOT_API_KEY=your-moonshot-api-key
+MOONSHOT_ENDPOINT=https://api.moonshot.ai/v1/chat/completions
+MOONSHOT_MODEL=kimi-k2-turbo-preview
+
+# ── Primary provider (Lynkr uses this when no tier matches) ───────────────────
+# Set to whichever you want as the default fallback
+MODEL_PROVIDER=azure-openai
+
+# ── Tier Routing ───────────────────────────────────────────────────────────────
+# SIMPLE   → Ollama local (free)
+# MEDIUM   → Moonshot Kimi (cheap, fast)
+# COMPLEX  → Azure OpenAI GPT-4o (powerful)
+# REASONING→ Azure OpenAI o3-mini (best reasoning)
+TIER_SIMPLE=ollama:minimax-m2.5:cloud
+TIER_MEDIUM=ollama:minimax-m2.5:cloud
+TIER_COMPLEX=moonshot:kimi-k2.6
+TIER_REASONING=azure-openai:gpt-5.2-chat
+
+# ── Token Optimisations (these are what LiteLLM/Portkey don't have) ────────────
+SMART_TOOL_SELECTION=true
+PROMPT_CACHE_ENABLED=true
+SEMANTIC_CACHE_ENABLED=true
+SEMANTIC_CACHE_THRESHOLD=0.95
+HISTORY_COMPRESSION_ENABLED=true
+TOOL_INJECTION_ENABLED=false
+
+# ── Optional: make routing decisions visible in responses ──────────────────────
+LYNKR_VISIBLE_ROUTING=true

package/benchmark-configs/portkey-config.json

@@ -0,0 +1,60 @@
+{
+  "_comment": "Portkey Gateway Config — multi-provider conditional routing",
+  "_note": "Portkey has NO automatic complexity detection. This config uses max_tokens as a proxy for complexity. For real tier routing pass x-portkey-metadata: { 'tier': 'simple|medium|complex|reasoning' } from your client.",
+
+  "strategy": {
+    "mode": "conditional"
+  },
+
+  "conditions": [
+    {
+      "_comment": "SIMPLE — short requests (max_tokens <= 256) → Ollama",
+      "condition": {
+        "query.max_tokens": { "$lte": 256 }
+      },
+      "target": {
+        "provider": "ollama",
+        "customHost": "http://localhost:11434",
+        "override_params": {
+          "model": "qwen2.5-coder:7b"
+        }
+      }
+    },
+    {
+      "_comment": "MEDIUM — metadata tier=medium → Moonshot",
+      "condition": {
+        "metadata.tier": { "$eq": "medium" }
+      },
+      "target": {
+        "provider": "openai",
+        "apiKey": "{{MOONSHOT_API_KEY}}",
+        "baseURL": "https://api.moonshot.ai/v1",
+        "override_params": {
+          "model": "moonshot-v1-8k"
+        }
+      }
+    },
+    {
+      "_comment": "REASONING — metadata tier=reasoning → Azure OpenAI o3-mini",
+      "condition": {
+        "metadata.tier": { "$eq": "reasoning" }
+      },
+      "target": {
+        "provider": "azure-openai",
+        "apiKey": "{{AZURE_OPENAI_API_KEY}}",
+        "resourceName": "YOUR-RESOURCE",
+        "deploymentId": "o3-mini",
+        "apiVersion": "2024-12-01-preview"
+      }
+    }
+  ],
+
+  "default": {
+    "_comment": "COMPLEX — everything else → Azure OpenAI GPT-4o",
+    "provider": "azure-openai",
+    "apiKey": "{{AZURE_OPENAI_API_KEY}}",
+    "resourceName": "YOUR-RESOURCE",
+    "deploymentId": "gpt-4o",
+    "apiVersion": "2024-08-01-preview"
+  }
+}