harnext is an open-source AI coding agent with harness engineering, built by FlowHunt. It is an interactive terminal agent that reads, writes, and edits files, runs shell commands, and drives MCP servers — and generates staged GitHub Actions workflows so AI can pick up GitHub issues and ship them, end to end.

How does harnext automate GitHub issues into pull requests?

harnext setup writes a staged pipeline of GitHub Actions workflows. Each stage is a label; promoting a label triggers the next stage. The pipeline moves through issue:tagged → stage:plan → stage:implement → stage:review → stage:merged, with a dedicated AI agent handling each step automatically.

Which AI models and providers does harnext support?

harnext supports 20+ LLM providers out of the box via pi-ai, including Anthropic, OpenAI, Google, Ollama, NVIDIA NIM, llama.cpp, vLLM, Mistral, and more. You can switch providers with a single --provider and --model flag, and even use a different provider per pipeline stage.

Can I use Claude Code or OpenAI Codex with harnext?

Yes. Each harnext pipeline stage can dispatch a different executor. You can use the harnext REPL for interactive work, and hand a CI stage off to claude-code or codex when running unattended — with approval and sandbox bypass flags applied automatically.

What tools does the harnext agent have?

harnext ships with a minimal, focused six-tool executor: read, write, edit, bash, skill, and MCP. That's the whole agent — no wrappers, no glue scripts, no hidden orchestration layer.

Can I run harnext with local models to keep my code private?

Yes. Point harnext at Ollama or any NVIDIA-compatible endpoint and your code never has to leave your machine. You can use the interactive REPL on a cloud provider and dispatch CI stages to a local Llama model — all with a single --provider flag.

How do I set up a self-hosted runner with harnext?

Run harnext setup in your repository and it will register a GitHub Actions self-hosted runner pinned to your repo with a project-specific label, install it as a systemd or launchd service, and keep it alive across reboots. harnext runner status and harnext runner logs tail the daemon.

What are harnext skills?

Skills are reusable agent bundles you attach to any pipeline stage. Built-in skills include review (code review), init (project initialisation), and browser-verify (visual regression checking). You can author your own and share them across repositories.

How does harnext connect to MCP servers?

harnext can connect to any MCP-compatible server in either proxy or direct mode. This lets the agent call browser automation tools, databases, search APIs, or any other MCP service as a native step in the pipeline.

Is harnext open source and how do I install it?

harnext is MIT licensed and open source. Install it globally with `npm install -g harnext` (Node.js ≥ 20 required). The source is at github.com/QualityUnit/harnext and the package is on npm.

Harnext by FlowHunt Learn more

AI coding agent. Harness-engineered.

harnext is the CI harness around your AI coding agent — it generates staged GitHub Actions workflows, label-driven promotion, and a self-hosted runner that carry an issue from triage to a merged PR, end to end.

Try Harnext now View on GitHub

harnext AI coding agent picking up a GitHub issue and opening a PR

harnext by FlowHunt

An open-source AI coding agent and CI harness that carries a GitHub issue from label to merged PR — autonomously. Built by the FlowHunt team, for developers who want AI in their workflow without giving up control.

LLM providers: 20+
Agent tools: 6
Pipeline stages: 5
Open source: MIT

Features

Everything you'd build into a coding agent — in one CLI.

harnext ships the agent, the harness, and the runner. No glue scripts, no agent server.

Interactive Terminal Agent: A REPL that reads, writes, edits, runs bash, and drives MCP servers. One-shot mode with -p for scripts and CI.
20+ Providers, One CLI: Anthropic, OpenAI, Google, Ollama, NVIDIA, and more via pi-ai. Switch with --provider and --model on the fly.
Harness for Your Repo: Generates GitHub Actions workflows that pick up issues by label, run the right stage, and post results back.
Self-Hosted Runner: Register a daemon on your own machine (systemd / launchd) and run stages on your hardware. Idempotent install.
Skills & MCP: Bundle reusable skills (review, init, browser-verify) and connect any MCP server with proxy or direct mode.
Replays & Status: harnext status shows active runs; replays let you re-run any session step-by-step. Sessions live in ~/.harnext.

The Agent

Minimal agent. Local models.

harnext is a small, focused CLI agent. Six tools, a few skills, and an MCP bridge — that's the whole executor. Point it at a local model and your codebase never leaves the machine.

harnext running with Ollama llama3.2 on localhost

Tiny Tool Surface.: read · write · edit · bash · skill · MCP. That's the whole agent. No wrappers, no glue scripts, no hidden orchestration.
Local-First Models.: Point harnext at Ollama or NVIDIA out of the box, or any OpenAI-compatible endpoint. Your code never has to leave your machine.
Switch Providers Per Stage.: Run interactive REPL on Anthropic, then dispatch a CI stage to a local Llama. One flag: --provider, --model.
Open & Inspectable.: MIT-licensed monorepo, plain-JSON sessions in ~/.harnext. Replay any run, audit any tool call.

harnext drives itself — or any agent you trust.

Each pipeline stage can dispatch a different executor. Use the harnext REPL for interactive work, and pick claude-code or codex per-stage when running unattended in CI.

harnext
Built-in interactive REPL. Read, write, edit, bash, and MCP — driven by your provider of choice (Anthropic, OpenAI, Google, Ollama, +20 more via pi-ai).
Learn more
Claude Code
Anthropic's CLI, dispatched. harnext hands a stage to claude-code with --max-turns and sandbox flags applied automatically.
Learn more
Codex
OpenAI's CLI, dispatched. Same interface as Claude Code — approval and sandbox bypass flags wired in for unattended runs.
Learn more

The Harness

From a GitHub issue to a merged PR — on rails.

harnext setup writes a staged pipeline of GitHub Actions workflows. Each stage is a label; promoting a label triggers the next stage. Run on GitHub-hosted runners, your self-hosted daemon, or both.

harnext five-stage pipeline with generated workflow YAML

issue:tagged: Tagger workflow applies the first stage label as soon as an issue is opened.
stage:plan: Agent reads the issue, writes a structured implementation plan back as a GitHub comment.
stage:implement: Agent opens a branch, edits files according to the plan, and raises a pull request.
stage:review: Reviewer agent reads the diff, posts inline comments, and triggers a fix workflow if changes are needed.
stage:merged: Post-merge cleanup workflow runs after the PR lands — closing issues, updating project state.

Three commands to a running agent.

harnext is designed to be productive in minutes. Install globally, run setup once per repository, then start your interactive agent.

1. Install
npm install -g harnext Requires Node.js ≥ 20. Stored at ~/.harnext.
View on npm
2. Set Up the Pipeline
cd my-repo && harnext setup Adds labels, writes the staged GitHub Actions workflows, and optionally registers a self-hosted runner.
Quick start guide
3. Use It as Your Coding Agent
harnext The same CLI is your interactive agent — read, write, run, drive MCP. Pass -p for one-shot use.
See the docs

Self-Hosted Runner

Run the agent on your own hardware.

harnext setup registers a GitHub Actions self-hosted runner pinned to your repo, installs it as a systemd or launchd service, and keeps it alive across reboots. harnext runner status and harnext runner logs tail the daemon.

Idempotent Install:: Re-running setup skips work that's already done — safe to run multiple times.
Project-Pinned Label:: A unique label prevents shared-host runners from picking up jobs meant for another project.
Public-Repo Guardrail:: harnext setup confirms fork-PR approval gates are on before registering a public-repo runner.
Clean Uninstall:: Best-effort cleanup on uninstall: deregister, stop service, drop all runner artifacts.

Frequently asked questions

: harnext is an open-source AI coding agent with harness engineering, built by FlowHunt. It is an interactive terminal agent that reads, writes, and edits files, runs shell commands, and drives MCP servers — and generates staged GitHub Actions workflows so AI can pick up GitHub issues and ship them, end to end.
: harnext setup writes a staged pipeline of GitHub Actions workflows. Each stage is a label; promoting a label triggers the next stage. The pipeline moves through issue:tagged → stage:plan → stage:implement → stage:review → stage:merged, with a dedicated AI agent handling each step automatically.
: harnext supports 20+ LLM providers out of the box via pi-ai, including Anthropic, OpenAI, Google, Ollama, NVIDIA NIM, llama.cpp, vLLM, Mistral, and more. You can switch providers with a single --provider and --model flag, and even use a different provider per pipeline stage.
: Yes. Each harnext pipeline stage can dispatch a different executor. You can use the harnext REPL for interactive work, and hand a CI stage off to claude-code or codex when running unattended — with approval and sandbox bypass flags applied automatically.
: harnext ships with a minimal, focused six-tool executor: read, write, edit, bash, skill, and MCP. That's the whole agent — no wrappers, no glue scripts, no hidden orchestration layer.
: Yes. Point harnext at Ollama or any NVIDIA-compatible endpoint and your code never has to leave your machine. You can use the interactive REPL on a cloud provider and dispatch CI stages to a local Llama model — all with a single --provider flag.
: Run harnext setup in your repository and it will register a GitHub Actions self-hosted runner pinned to your repo with a project-specific label, install it as a systemd or launchd service, and keep it alive across reboots. harnext runner status and harnext runner logs tail the daemon.
: Skills are reusable agent bundles you attach to any pipeline stage. Built-in skills include review (code review), init (project initialisation), and browser-verify (visual regression checking). You can author your own and share them across repositories.
: harnext can connect to any MCP-compatible server in either proxy or direct mode. This lets the agent call browser automation tools, databases, search APIs, or any other MCP service as a native step in the pipeline.
: harnext is MIT licensed and open source. Install it globally with `npm install -g harnext` (Node.js ≥ 20 required). The source is at github.com/QualityUnit/harnext and the package is on npm.

Learn more

GitMCP

Integrate FlowHunt with GitMCP to provide AI assistants with instant access to up-to-date GitHub documentation and code, reducing hallucinations and boosting co...

Aug 12, 2025 4 min read

AI GitMCP +4

Heroku MCP

Integrate FlowHunt with the Heroku MCP Server for AI-powered management of Heroku resources. Automate app deployment, dyno scaling, add-on provisioning, and Pos...

Aug 12, 2025 5 min read

AI Heroku +4