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

[codegen-sh]

Summary

Adds a local proxy server that auto-launches when eversale starts and routes all LLM calls to your configured backend. This eliminates the need to modify eversale's core logic for different LLM providers.

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    │     └─────────────────┘
└──────────────┘     └──────────────────────┘

New Files

File	Lines	Description
proxy_server.py	~620	aiohttp gateway — 4 backend adapters (Anthropic, OpenAI, Ollama, Custom), full SSE streaming, format translation
proxy_launcher.py	~250	Subprocess lifecycle — idempotent start, PID management, health-check gating, CLI
proxy_config.py	~165	Config hub — backend selection, model mapping, env var routing

Proxy Endpoints

Endpoint	Method	Description
/v1/chat/completions	POST	OpenAI-compatible chat (streaming + non-streaming)
/v1/models	GET	Model listing
/api/chat	POST	Ollama-compatible chat
/health	GET	Readiness probe

Bug Fixes (PR #213)

• llm_fallback_chain.py: Added missing import os (was causing NameError at import time)
• llm_fallback_chain.py: Added missing _looks_like_ollama_endpoint() method (was causing AttributeError at runtime)

Modified Files

• gpu_llm_client.py: Default URL → localhost:8765, added _is_local_proxy(), proxy handles auth
• config.yaml: Added proxy section, URLs point to proxy
• .env.example: Full documentation of proxy env vars
• README.md: Architecture diagram, quick-start for all 4 backends

Quick Start

export LLM_BACKEND=anthropic  # or openai, ollama, custom
export ANTHROPIC_API_KEY=sk-ant-...
python proxy_launcher.py start

Based on PR #213 (codegen-bot/eversale-local-api-integration-a7f3c2).

---

💻 View my work • 👤 Initiated by @Zeeeepa • About Codegen
⛔ Remove Codegen from PR • 🚫 Ban action checks

---

Summary by cubic

Adds an auto-launching local proxy that routes all LLM calls to your chosen backend (Anthropic/OpenAI/Ollama/custom), removing provider-specific logic from eversale and simplifying setup. Default LLM requests now go through http://127.0.0.1:8765 with streaming and model mapping handled by the proxy.

• New Features

  • Added proxy_server.py: OpenAI-compatible /v1/chat/completions, /v1/models, and Ollama-compatible /api/chat with SSE streaming; adapters for Anthropic, OpenAI, Ollama, and custom.
  • Added proxy_launcher.py: idempotent background start, health-check gating, PID management; binds to 127.0.0.1:8765.
  • Updated gpu_llm_client.py: defaults to the local proxy; auth handled by proxy; supports proxy detection.
  • Added proxy_config.py, config.yaml, .env.example, and README.md: backend selection via LLM_BACKEND, model mapping, and quick start.

• Bug Fixes

  • llm_fallback_chain.py: added missing import os and _looks_like_ollama_endpoint() to fix NameError/AttributeError.

^{Written for commit bcd8193. Summary will update on new commits.}