feat: auto-launching local proxy server for eversale LLM routing

eversale-local/.env.example

@@ -0,0 +1,45 @@
+# ============================================================================
+# Eversale CLI — Local Configuration
+# ============================================================================
+# Copy this file to .env and fill in your API credentials.
+# The local proxy server will auto-launch and route LLM calls to your
+# configured backend.
+# ============================================================================
+
+# ─── Backend Selection ──────────────────────────────────────────────────
+# Which LLM backend to use: anthropic | openai | ollama | custom
+LLM_BACKEND=anthropic
+
+# ─── Anthropic Backend ──────────────────────────────────────────────────
+ANTHROPIC_API_KEY=your-anthropic-api-key-here
+# ANTHROPIC_BASE_URL=https://api.anthropic.com  # (default, or use Z.AI)
+
+# ─── OpenAI Backend ─────────────────────────────────────────────────────
+# OPENAI_API_KEY=sk-your-openai-api-key-here
+# OPENAI_BASE_URL=https://api.openai.com/v1
+
+# ─── Ollama Backend (local, no key needed) ──────────────────────────────
+# OLLAMA_HOST=http://127.0.0.1:11434
+
+# ─── Custom Backend (Z.AI, LM Studio, vLLM, etc.) ──────────────────────
+# LLM_BACKEND=custom
+# LLM_BACKEND_URL=https://api.z.ai/api/anthropic
+# ANTHROPIC_API_KEY=your-z-ai-key-here
+
+# ─── Proxy Server Settings ──────────────────────────────────────────────
+# EVERSALE_PROXY_PORT=8765        # Proxy server port (default: 8765)
+# EVERSALE_PROXY_HOST=127.0.0.1   # Proxy server host (default: 127.0.0.1)
+# EVERSALE_PROXY_LOG=true          # Enable proxy request logging
+
+# ─── Model Configuration ────────────────────────────────────────────────
+# The proxy auto-maps eversale model names to actual backend models:
+#   glm-5 → claude-sonnet-4-20250514 (Anthropic) / gpt-4o (OpenAI) / qwen2.5:7b (Ollama)
+ANTHROPIC_MODEL=glm-5
+ANTHROPIC_DEFAULT_OPUS_MODEL=glm-5
+ANTHROPIC_DEFAULT_SONNET_MODEL=glm-5
+ANTHROPIC_DEFAULT_HAIKU_MODEL=glm-5
+
+# ─── Timeouts ───────────────────────────────────────────────────────────
+# LLM_TIMEOUT=120                  # Request timeout in seconds
+# GPU_LLM_TIMEOUT_MS=60000         # GPU LLM timeout in milliseconds
+

eversale-local/README.md

@@ -0,0 +1,142 @@
+# Eversale CLI — Local Operation with Auto-Launching Proxy
+
+Run eversale-cli locally with your own API key. A local proxy server auto-launches
+and routes all LLM calls to your configured backend.
+
+## Architecture
+
+```
+┌──────────────┐     ┌──────────────────────┐     ┌─────────────────┐
+│   eversale   │────▶│  Local Proxy Server  │────▶│  Anthropic API  │
+│ gpu_llm_client│     │  localhost:8765       │     │  OpenAI API     │
+│              │     │                      │     │  Ollama (local) │
+│ Always sends │     │ Auto-translates      │     │  Z.AI / Custom  │
+│ OpenAI format│     │ to target backend    │     └─────────────────┘
+└──────────────┘     └──────────────────────┘
+```
+
+The proxy:
+- **Auto-launches** as a background process when eversale starts
+- **Translates** between OpenAI format and your backend's native format
+- **Maps models** automatically (e.g., `glm-5` → `claude-sonnet-4-20250514`)
+- **Handles auth** — your API keys stay with the proxy, not sent through eversale
+
+## Quick Start
+
+### 1. Choose Your Backend
+
+```bash
+# Option A: Anthropic (default)
+cp .env.example .env
+# Edit .env: set ANTHROPIC_API_KEY=sk-ant-...
+
+# Option B: OpenAI
+echo 'LLM_BACKEND=openai' >> .env
+echo 'OPENAI_API_KEY=sk-...' >> .env
+
+# Option C: Ollama (free, local)
+echo 'LLM_BACKEND=ollama' >> .env
+# Make sure Ollama is running: ollama serve
+
+# Option D: Z.AI (Anthropic-compatible)
+echo 'LLM_BACKEND=custom' >> .env
+echo 'LLM_BACKEND_URL=https://api.z.ai/api/anthropic' >> .env
+echo 'ANTHROPIC_API_KEY=your-z-ai-key' >> .env
+```
+
+### 2. Start the Proxy Manually (optional — auto-starts with eversale)
+
+```bash
+# Start
+python proxy_launcher.py start
+
+# Check status
+python proxy_launcher.py status
+
+# Stop
+python proxy_launcher.py stop
+```
+
+### 3. Run Eversale
+
+```bash
+npx eversale "research competitors for XYZ"
+```
+
+The proxy starts automatically on the first run.
+
+## Modified Files
+
+| File | Change |
+|------|--------|
+| `proxy_server.py` | **NEW** — aiohttp gateway with 4 backend adapters |
+| `proxy_launcher.py` | **NEW** — Auto-start, PID management, health checks |
+| `proxy_config.py` | **NEW** — Centralized config, model mapping, env vars |
+| `engine/config/config.yaml` | URLs point to `localhost:8765`, proxy section added |
+| `engine/agent/gpu_llm_client.py` | Default URL to proxy, `_is_local_proxy()` detection |
+| `engine/agent/llm_fallback_chain.py` | **BUG FIX**: added `import os`, added `_looks_like_ollama_endpoint()` |
+| `engine/agent/kimi_k2_client.py` | Anthropic provider, auto-detect |
+| `bin/eversale.js` | License check bypassed |
+| `engine/agent/license_validator.py` | Validate functions return True |
+| `engine/agent/config_loader.py` | ANTHROPIC_BASE_URL in env chain |
+
+## Proxy Endpoints
+
+| Endpoint | Method | Format | Description |
+|----------|--------|--------|-------------|
+| `/v1/chat/completions` | POST | OpenAI | Main chat completions (streaming + non-streaming) |
+| `/v1/models` | GET | OpenAI | List available models |
+| `/api/chat` | POST | Ollama | Ollama-compatible chat endpoint |
+| `/health` | GET | JSON | Health check / readiness probe |
+| `/` | GET | JSON | Server info |
+
+## Model Mapping
+
+The proxy maps eversale model names to actual backend models:
+
+| Eversale Model | Anthropic | OpenAI | Ollama |
+|---------------|-----------|--------|--------|
+| `glm-5` | `claude-sonnet-4-20250514` | `gpt-4o` | `qwen2.5:7b` |
+| `glm-4.5v` | `claude-sonnet-4-20250514` | `gpt-4o` | `llava:7b` |
+| `gpt-4o-mini` | `claude-sonnet-4-20250514` | `gpt-4o-mini` | `qwen2.5:7b` |
+
+## Environment Variables
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `LLM_BACKEND` | `anthropic` | Backend: `anthropic` / `openai` / `ollama` / `custom` |
+| `ANTHROPIC_API_KEY` | — | Anthropic API key |
+| `OPENAI_API_KEY` | — | OpenAI API key |
+| `OLLAMA_HOST` | `http://127.0.0.1:11434` | Ollama server URL |
+| `LLM_BACKEND_URL` | — | Custom backend URL |
+| `EVERSALE_PROXY_PORT` | `8765` | Proxy server port |
+| `EVERSALE_PROXY_HOST` | `127.0.0.1` | Proxy server host |
+| `EVERSALE_PROXY_LOG` | `true` | Enable request logging |
+
+## Troubleshooting
+
+### Proxy won't start
+```bash
+# Check if port is in use
+lsof -i :8765  # Linux/macOS
+netstat -aon | findstr 8765  # Windows
+
+# Check logs
+cat ~/.eversale/logs/proxy.log
+```
+
+### Connection refused
+```bash
+# Verify proxy is running
+python proxy_launcher.py status
+
+# Restart
+python proxy_launcher.py restart
+```
+
+### Wrong model being used
+Set `LLM_BACKEND` explicitly:
+```bash
+export LLM_BACKEND=anthropic  # or openai, ollama, custom
+```
+