Installation and LLM Setup

Install

pip install git+https://github.com/arunsoman/atomic-forge.git   # one line, no checkout needed
atomic-forge --help                                            # sanity check the CLI

Or from a checkout: pip install -e ".[dev]" (also installs pytest for the suite). Requires Python ≥3.10. At runtime forge needs an OpenAI-compatible LLM endpoint — point it at OpenAI, or a local Ollama model that supports tool-calling.

LLM configuration

default_llm() resolves, in order:

1. FORGE_MOCK=1 — use your own zero-network mock (register one via atomic_forge.llm.set_mock_factory(...)), useful for demos and CI.
2. FORGE_API_KEY / FORGE_BASE_URL / FORGE_MODEL — any OpenAI-compatible endpoint: OpenAI itself, a local vLLM/llama.cpp/Ollama proxy, OpenRouter, a corporate gateway.
3. OPENAI_API_KEY (+ optional OPENAI_BASE_URL/OPENAI_MODEL) — the common case where you already have this set.
4. Otherwise: raises with a message naming exactly what to set. Never silently falls back to a fake key against real api.openai.com.

# OpenAI
export FORGE_API_KEY=sk-... FORGE_BASE_URL=https://api.openai.com/v1 FORGE_MODEL=gpt-4o-mini

# Local Ollama (tool-calling model required)
export FORGE_MODEL=qwen3.5:cloud FORGE_BASE_URL=http://localhost:11434/v1 FORGE_API_KEY=ollama

Privacy: nothing has to leave your machine

--local-only refuses to run against a non-loopback/private LLM endpoint — it enforces the "nothing leaves this machine" claim instead of merely permitting it. See Data-Privacy-No-Training. Forge never trains on your code; there is no telemetry.

Optional companion: CIE

forge's repair loop can use CIE — the Code Insight Engine as its code-graph backend, served as a real MCP server over stdio (the same surface Claude Code / Cursor consume):

pip install git+https://github.com/arunsoman/cie.git pytest

CIE is required for the fix pipeline (Issue-to-PR) and optional everywhere else (CIE-Integration).