A coding agent that runs on your own models
AgenC Core is an open source coding agent for your terminal. Point it at Ollama, LM Studio or any OpenAI-compatible server and it works with no API key and no cloud account. Your prompts go to your own model instead of a cloud API.
With Ollama
ollama pull qwen3-coder:30b curl -fsSL https://get.agenc.ag/install.sh | sh agenc --provider ollama --model qwen3-coder:30b
AgenC talks to Ollama at localhost:11434. If Ollama runs on another machine, set OLLAMA_BASE_URL, for example http://gpu-box:11434.
With LM Studio
Load a model in LM Studio and start its local server, on port 1234 by default. Then start AgenC with that model's ID:
agenc --provider lmstudio --model <model-id>
With llama.cpp, vLLM or another OpenAI-compatible server
export OPENAI_COMPATIBLE_BASE_URL=http://localhost:8000/v1 agenc --provider openai-compatible --model <model-name>
The context window, handled
AgenC asks the server how much context the model has: Ollama through /api/show, llama.cpp through the context size it was started with, vLLM through max_model_len. It plans the conversation around that number, so a model served with a small window does not get overfilled.
Picking a model
Choose a model that supports tool calling. The agent works by calling tools: reading files, editing them and running commands. Coding models such as Qwen3 Coder are a good place to start.
Prefer a hosted open model?
Ollama Cloud has its own provider. Set OLLAMA_API_KEY and start agenc --provider ollama-cloud. Its default model is DeepSeek V4.1 Flash. All 25 providers
Questions
Can I use AgenC Core with Ollama?
Yes. Start it with agenc --provider ollama --model followed by the model name. It connects to Ollama at localhost:11434 and needs no API key. Set OLLAMA_BASE_URL if Ollama runs on another machine.
Does AgenC Core work with LM Studio?
Yes. Start LM Studio's local server, then run agenc --provider lmstudio --model with the model's ID.
Can I use llama.cpp or vLLM?
Yes. Set OPENAI_COMPATIBLE_BASE_URL to the server's /v1 address and start agenc --provider openai-compatible. AgenC reads the served context size from the server.
Which local models work best?
Models that support tool calling, because the agent reads files, edits them and runs commands through tools. Coding models such as Qwen3 Coder are a good start.
Is it free to run?
AgenC Core is open source under the MIT license, and a model on your own machine costs nothing per token.
Open source under the MIT license. Source code · All providers · Verify your download