GitHub

Embeddable Go Agent Runtime

Local-model-first via LM Studio. In-process tool-calling LLM loop
for build orchestrators and CI systems.

agent.Run(ctx, request) — no subprocess overhead, direct state sharing with the host.

LM Studio and Ollama support via OpenAI-compatible API. Zero marginal cost for routine tasks.

Every LLM turn and tool call logged to JSONL. Sessions are replayable and cost-tracked.

Shipped tools — read, write, edit, bash, glob, grep, ls, patch, task. Simple, auditable, benchmark-ready.

OpenAI-compatible (LM Studio, Ollama, OpenAI, Azure) and Anthropic Claude. One interface.

ddx-agent -p ‘prompt’ proves the library works. Config, session logs, replay, usage reporting.