Mingbird Agent
github.com/mingbird/mingbird-agent- Category
- AI Agents
- Rank
- No. 3625Tools index
Previous survey · No. 3482 ·
- Type
- AGENT
- GitHub
- 27 stars
- Latest release
- v3.1.11
- Date
About
Mingbird is a local-first agent harness for Windows and Ollama aimed at 2 to 9B parameter models. Its repository reports that the same 2B model scores 0.017 to 0.821 across four agent harnesses on a 288-cell public benchmark. It includes a one-click mode that blocks outbound connections and has no telemetry.
At a glance
Mingbird runs a model through an agent that can edit files, run commands and feed test errors back into another attempt. It is worth evaluating if you want to see whether a small local model can finish a bounded task on your own hardware.
Good fit
Developers who can review the source and dependencies and test the agent in an isolated environment with a disposable workspace.
Consider something else when
Skip it if you need a signed Windows installer or cannot isolate the agent from private files and credentials. The documented Windows installer is unsigned.
Start here
Review the source setup and use a disposable environment. Select a local inference engine, enable the documented offline mode, and try adding a test to a synthetic project. Inspect the diff and run the test independently.
Cost and requirements
The Apache 2.0 project documents a Python source setup and local engines including Ollama. Model memory and runtime depend on the model and hardware. Optional cloud inference has separate provider costs and data handling.
Limits to know
The agent runs commands and writes files. The documented offline mode disables web tools and cloud inference within Mingbird. Use operating-system isolation to constrain network and file access during evaluation.
Sources checked 2026-11-05. Source-based assessment.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.