Can I use OpenACP with local/self-hosted models (Ollama, vLLM)?

[amanbuild]

I run Ollama locally with Llama 3 and DeepSeek Coder. Can I route these through OpenACP instead of paying for API keys? Privacy is a concern for my client's codebase.

[amanbuild]

OpenACP connects to agents, not models directly. The model choice depends on which agent you use:

• Aider supports Ollama and any OpenAI-compatible endpoint (vLLM, LM Studio, text-generation-webui). Point it at localhost:11434 and it works through OpenACP.
• Claude Code and Codex CLI are locked to their respective providers (Anthropic, OpenAI).
• goose (by Block) supports local models via OpenAI-compatible APIs.
• Cline supports local models through its configuration.

So yes — use an agent that supports local models, and OpenACP will bridge it to your chat platform like any other agent. The privacy benefit is real: your code never leaves your machine.

Performance note: local models on consumer hardware are noticeably slower than cloud APIs for complex coding tasks. A 70B model on a 4090 is usable but not fast. For privacy-sensitive work, it's a worthy tradeoff.

[romgX]

Yes, you can! Ollama exposes an OpenAI-compatible API endpoint (usually http://localhost:11434), so you can point OpenACP to it directly.

Quick setup:

1. Make sure Ollama is running locally
2. In OpenACP config, add your local endpoint as a provider
3. Set the model names to match what you have in Ollama (e.g., llama3, deepseek-coder)

For privacy-sensitive projects, this is actually a solid approach — everything stays on your machine, no data leaves your network. Just keep in mind local models might be slower than cloud APIs depending on your hardware.

One thing to watch: context window size. DeepSeek Coder handles longer contexts better than Llama 3 for code tasks, so you might want to route different request types to different models based on that.