Apache-2.0 · Open source · Loopback studio

Message teammates.
Not another chat thread.

Open Coding Partner ships a local studio where teammates keep identity and memory, draft a plan before they touch your repo, run behind explicit approvals, and hand work to QA — with a virtual computer you can watch or drive.

OpenCodingPartner · Studio 127.0.0.1:4317
Add validation tests and queue QA when the suite passes.
On it. I will draft a plan first — no edits until you approve.
Tell your teammates what to build…

Illustrative studio chrome — the product runs locally on your machine.

One workspace.
Everyone on the thread.

Coding, review, and security share one project room. You will see who is working, what changed, and when a gate needs you — from a roster and audit trail we keep on your machine, not in a hosted inbox.

  • $0 core forever
  • Plan before code
  • QA handoffs built in
Roster Live
  • OCP CoderImplementing · 3 files
  • QA ReviewerReading diff
  • SecurityAwaiting approval
OCP CoderImplementing · 3 files
OpenCodingPartner / validate

What we ship

Decide, then act

Plans and consequential tools wait on Allow once, Deny, or Redirect. Unattended runs fail closed — we record every outcome on the audit page.

A computer per teammate

Isolated displays and browser profiles per bot. Take control for sign-in, return responsibility when you are done — already in the studio you clone today.

Your models, your meter

No model in the box. OpenRouter or any OpenAI-compatible endpoint, including local Ollama. Keys never land in task logs or state files.

Work with many teammates at once

We built the studio for a lead plus coding, QA, security, and computer roles in one workspace — with routines, skills, channel drafts, and an usage cap you control.

Roster-first studio

Scan the sidebar like a team directory — not a graveyard of old chats.

Memory

Per-workspace notes that survive the next task.

History recall

Search prior transcripts in the same repo.

Guarded tools

Bounded reads/writes with secret scanning.

Provider-neutral

No default model. Route through OpenRouter or any OpenAI-compatible server — including local Ollama and LM Studio.

Give each teammate a job

Same studio, different responsibilities — pick a role to see how OCP frames the work.

OCP Coder

Ships features end to end: reads the repo, writes a plan, waits for your approve, edits files, runs checks, and opens a PR-ready diff.

A computer you can share

Each teammate can drive an isolated display — browser, mouse, keyboard — while you watch a live stream, click to take control, then return responsibility.

  • 1280×800 virtual desktop per workspace
  • Human takeover without stopping the studio
  • Isolated browser profiles between bots on Linux
MJPEG · LIVE

What it costs

$0

The core is free and open source. You pay only the model providers you choose — same as bringing your own keys to any serious agent stack.

Questions

No. The studio binds to loopback (127.0.0.1) by default. This marketing site is separate; the product runs on your machine.

Any model you configure — OpenRouter, or an OpenAI-compatible endpoint (Ollama, LM Studio, vLLM, custom). There is no default model baked in.

Local state under your OS data directory (OCP_DATA_DIR optional). API keys are not written to task logs or backups.

Consequential tool use requires explicit approval (allow once or deny). Plans wait for your approve before edits when partner rules apply.

Yes — Apache-2.0. Source on GitHub. Upstream runtimes are credited in NOTICE; no proprietary vendor source is vendored.

Get Open Coding Partner

Requires Node.js ≥ 22.14 and pnpm. Docker optional for container isolation.

Quick install
git clone https://github.com/bgagandeep/OpenCodingPartner.git
cd OpenCodingPartner
corepack enable && corepack pnpm install && corepack pnpm build
export OPENROUTER_API_KEY="your-key"
corepack pnpm app
open http://127.0.0.1:4317

Onboarding guide · GitHub