mini-agent

Local-First
AI CLI.

Local-first AI for coding, goal execution, and release discipline. Built with Go, Ollama, and SQLite.

mini-agent — goal mode
❯ mini-agent /goal "Stabilize the MCP execution loop and validate the repo"
Analyzing repository context...
description internal/mcp/server.go Loaded (42ms)
description internal/goal/manager.go Loaded (18ms)
Generating patch proposal...
✔ Patch generated. Running validation tests...
Waiting for validation...
memory

Context-Aware Autopilot

Ranks likely-relevant files before making changes. It uses repository context, recent prompts, and summaries to start in the right place.

Ranked Context
  • 98% internal/mcp/server.go
  • 84% internal/agent/agent.go
  • 62% internal/planner/prompt.go
lock

Local-First Privacy

Built for local-first workflows, with Ollama as the default runtime for on-machine inference.

new_releases

Release Discipline

Tagged releases with changelog discipline and reproducible GitHub Actions distribution.

git tagv0.0.1

How it Works

search

1. Query

Define your goal. The agent maps the required context locally.

edit_document

2. Patch

Apply focused edits across the files that matter to the current goal.

fact_check

3. Validate

Tests are run to ensure the patch resolves the goal cleanly.