Open source · Desktop workbench

The desktop workspace where software gets built.

Brainstorm, code, run shell commands, test web apps in an inner browser, and review pull requests without leaving the window. Pi is bundled out of the box.

Works with your installed CLI agents

  • Pi
  • Codex CLI
  • Claude Code
  • OpenCode
  • Cline
  • Antigravity
  • Muse Code
  • +∞
DeepSeek v4 flash running inside CodeInOven to test and operate a local desktop app.

The All-Day Workbench

Spend your workday in one window.

No more bouncing across terminal windows, browser tabs, GitHub PRs, and chat boxes. Everything you need to build software sits in one workstation.

Inner browser

Test localhost, inspect DOM elements, and let agents test web flows directly in the window.

Attached terminal

A real shell in your project root. Run builds, migrations, and scripts next to your agent.

GitHub PR reviews

Review PRs, inspect diffs, resolve conflicts, and commit staged files without switching to a browser.

Ideation & prototyping

Draft PRDs, prototype quick spikes, and validate architectures before editing production code.

Telemetry & cost tracking

Track token burn, latency, and costs per step. Save repetitive build commands as reusable actions.

Projects & scopes

Switch between repositories instantly. Track work across Pinned, Todo, Spec, and Done.

Universal harness support

Runs bundled Pi, or connects to Claude Code, OpenCode, Codex, Cline, Ollama, and LM Studio.

Vision & desktop driver

Feed UI mockups and screenshots directly to models. Agents can drive the desktop UI to verify results.

Remote phone review

Pair via QR code. Check diffs, follow progress, and approve tool permissions from your phone.

Engineering Toolbox stages: Brainstorm, PRD, Spec, Assignment, Achievement, and Auto Pilot
Engineering toolbox: Pick the phase you need, from initial brainstorm and PRD to spec creation and implementation.
Scopes board organizing threads into Pinned, Todo, Spec, and Done columns across project scopes
Scopes board: Track work across Pinned, Todo, Spec, and Done columns for each project repository.
The CodeInOven composer, showing permission scope, model selector, and run controls.
Permission controls sit right in the composer, so you verify access before sending a prompt.

Zero Ceremony

No AGENTS.md or CLAUDE.md required.

Built-in instructions guide models to inspect code, adhere to repository conventions, write focused diffs, and test before finishing.

Built-in rules

Engineered discipline

Agents read files before editing, match existing style, keep diffs small, and test before finishing.

Checkpoints

No prompt loops

Work runs in phases with clear diffs and approvals. No endless prompt babysitting.

Instant start

Open and build

Open any project folder and start coding right away. Zero manual configuration.

Harnesses & Models

Bundled with Pi. Works with your existing tools.

CodeInOven comes bundled with Pi. Bring an API key or point to a local model. If you already have Claude Code, OpenCode, Codex, or Cline installed, CodeInOven runs them too.

Bundled

Pi is included. No CLI installs required. Add your model key or endpoint and start building.

Local models and private clouds

Supports OpenAI-compatible and Anthropic-compatible endpoints. Connect Ollama, LM Studio, vLLM, or dedicated private cloud servers with a base URL and model name.

  • Ollama http://localhost:11434/v1
  • LM Studio http://localhost:1234/v1
  • llama.cpp http://localhost:8080/v1

Hosted provider keys

Sign in through the provider or bring your own API key. Keys stay in local configuration files on your machine. CodeInOven never proxies your code or prompts through external middleman servers.

  • openai
  • anthropic
  • google
  • xai
  • deepseek
  • qwen
  • mistral
  • groq
  • openrouter
  • huggingface

The lifecycle

Spec first. Code after approval.

Solid engineering follows a reliable sequence: specify, review, approve, implement. Every step leaves a clear record you can inspect at any time.

  1. Specify

    A spec with fixed sections.

    Problem, Resolution, Success Criteria, Test Strategy, Commit Pattern, Constraints & Risks. Same structure every time, so nothing important stays unsaid.

    problemresolutionsuccess criteriatest strategycommit pattern
  2. Review

    Read it before anything runs.

    The agent drafts the spec from your thread. Annotate any section and send it back. Once you approve, that text is the contract the work is held to.

    annotationsversionsapprovesend back
  3. Approve

    Permissions are explicit.

    Tool calls ask within the tier you set. Auto Review by default; full access only if you switch it on yourself.

    Auto ReviewFull Access
  4. Implement

    Checked work, phase by phase.

    Each phase lands as a checkpoint with a diff. An audit pass tests the result against Success Criteria before you merge.

    checkpointsdiffsauditrollback
A CodeInOven run in progress showing reasoning traces, shell commands, and file reads with status
Reasoning steps, shell commands, execution time, and output stay visible on screen.

Our approach

We build what people use. If something is missing, send a PR.

No gimmicks or bloat. We build what developers actually need to ship code. If a harness, provider, or tool you use is missing, open an issue or bring a PR.

Built for all-day focus

Organized to keep you in flow without jumping between windows or losing context.

Driven by developer PRs

New harnesses and providers get added as the community needs them.

Local and private

Your code never routes through third-party proxies. State stays on your machine.

Principles

The rules it runs by

Determinism over vibes

Every run replays from saved state. If one cannot, that is a bug.

The human reviews

The agent proposes and executes. Nothing lands without your approval.

Never touch what is not yours

CodeInOven keeps its state in its own config directory and writes nothing uninvited into your repository.

Bounded on purpose

Threads, history chunks, and checkpoints are capped. Growth is a decision, not drift.

Your entire software day, in one place.

Free for personal and non-commercial use. Every run stays reviewable even after you close the thread.