Try it
npx @bitfieldcreek/bf-agent "summarize README.md"Pinned result
On 25 Oct 2026, bf-agent 1.1.25 scored 96 out of 100 on 34 graded tasks
(69 minutes). Smaller tasks ran on local qwen3-coder:30b. Harder builds ran on
kimi-k2.7-code:cloud. Windows, 32 cores, 64 GB RAM.
| What we asked | Tasks | Score | Result |
|---|---|---|---|
| Answer a question, edit one file, or add a small module | 8 | 99 | All passed |
| Fix a bug or add one behavior | 6 | 97 | All passed |
| Write a small command-line tool | 5 | 92 | All passed |
| Change a program that is already there | 5 | 97 | All passed |
| Build a larger tool from scratch | 10 | 94 | All passed |
| Do not change files you were only asked to read | 6 | — | 5 of 6 obeyed |
The miss: one read-only task also created README.md and src/index.mjs in
its throwaway workspace. Scores above count only whether the requested work was done.
Pairs with Aura Wire AI Command Center
Bitfield Ollama Agent
Ollama-based agent loop — free standalone, and built into Aura Wire AI Command Center as an agent runtime
Why use this?
Bitfield Ollama Agent turns Ollama into a multi-step agent: plan, call tools, observe results, and repeat until the task is done — all on your machine. On local Ollama models it costs nothing. It also ships inside Aura Wire AI Command Center as a built-in agent runtime.
CLI for one-off tasks, an embeddable SDK, and serve mode that exposes the agent loop behind an OpenAI-compatible API. Install globally with npm — or try once with npx. Inside Aura Wire AI Command Center, the Bitfield Ollama Agent runtime uses this same package (as an SDK sidecar, or the bf-agent CLI on your PATH); any other HTTP client can talk to serve mode with no Bitfield Ollama Agent–specific integration.
- Install with npm — npm install -g @bitfieldcreek/bf-agent — or run once via npx
- Runs the full agent loop on local Ollama — tools, retries, and sessions stay on your machine
- Built into Aura Wire AI Command Center — a ready-made agent runtime under Custom providers; Router’s Delegate to agent action can hand tasks to it without changing your IDE’s backend
- Free on local models; route through Wire to reach guarded, compressed cloud backends when you need them
- OpenAI-compatible /v1 server (serve mode) — for any HTTP client, or as an alternate Wire custom-provider destination
- CLI + SDK surfaces — script tasks or embed the loop in other software
Requirements
- Node.js 20+
- Ollama running locally (default http://127.0.0.1:11434)
- A tool-capable model installed (e.g. llama3.2, qwen3-coder)
- Aura Wire AI Command Center optional — the Bitfield Ollama Agent runtime is already on its Custom providers list
Install
- Install Ollama and pull a tool-capable model.
- Install Bitfield Ollama Agent: npm install -g @bitfieldcreek/bf-agent
- Run a quick task: bf-agent "what time is it right now?" (or npx for a one-off try)
- For an OpenAI-compatible endpoint: bf-agent serve --port 11435
- Using Aura Wire AI Command Center? Open Pipeline → Aura Router → Custom providers, enable the Bitfield Ollama Agent runtime (SDK sidecar or CLI), and add a Router rule that delegates agentic traffic to it.
- Air-gapped? Download the .zip below and use the bin/ launchers instead.