Llama App adds official agent integrations: run Pi and Codex with local models, fully offline
mervenoyann · x · 2026-11-08
The llama.cpp team and Hugging Face published official docs for connecting coding agents to Llama App, which serves local open models via an OpenAI-compatible endpoint (default http://127.0.0.1:9931/v1), keeping everything on-device and working fully offline.
How it works
- Llama App runs a local llama.cpp server; the agent sends prompts to it and executes file/terminal actions.
- Models with ≥32k context are recommended, ideally those flagged for "agentic coding".
Supported integrations
- Pi: install via curl, run /login llama.cpp to point at the server, then /model to pick a model. Pi is the harness behind OpenClaw.
- Codex CLI / ChatGPT Desktop: install Codex via npm and set the local baseurl and model in /.codex/config.toml.
The graphic also lists OpenClaw (OpenAI-compatible API) and Hermes (setup during onboarding). The project has 133.4K GitHub stars.
More from coding & agent
- Baseten Loops goes GA with serverless SFT/RL training, per-token pricing — baseten · 2026-11-08
- Software engineer asks how to bring AI agents into his family's retail business — FIREATWlLL · 2026-11-08
- Where do your agents run when the laptop closes? 4 hosting options compared — michaelTM_ai · 2026-11-08
- Where do browser agents actually break on the real web? Login walls, popups, or worse — lucy_yarntra007 · 2026-11-08
- Anthropic's 40-min AI-native SDLC playbook: code is no longer the bottleneck — blaizedsouza · 2026-11-08
- The software factory is the next extinct wrapper: coding agents will absorb orchestration value in 3-6 months — blaizedsouza · 2026-11-08