A developer’s answer to working with AI agents

Free to self-host · Built in Rust · Open code

AI agents say done.
rightcommit proves it.

I’m a developer. I build software with AI agents like Claude Code, Codex and others. AI agents have transformed how I code, but managing the entire development lifecycle remains a challenge. I wanted one simple workspace to go from idea to production-ready software, with agents doing the heavy lifting. Every step visible, quality verified, costs transparent, and me always in control. I couldn’t find it. So I built it. It’s called rightcommit.

Run it locally → Explore the workflow
On your machine or server Your keys, your models One container Blasting fast

Use Cases

These are the use cases I initially built rightcommit for.

rightcommit / Landing demo · Feature SDD
› Workflow ready · waiting for youManager draft · illustrative playback
step order which agent works go back and fix (with a limit)
Benefits

Build with confidence.
Stay in control.

Six ways rightcommit helps you turn AI-assisted work into verified results.

Verified outcomes

Move forward with test results and independent reviews you can inspect.

How it worksWork moves on only with proof: task boxes must be ticked, the server runs the tests, and a second agent, even a different one, reviews the work. If anything changes after the tests, tests and review run again.
Related discussion on Reddit ↗

Traceable specifications and code

Follow how each step turns a specification into files, changes and reviewable evidence.

How it worksEvery step saves its files as versions with a fingerprint (SHA-256) and line-by-line differences. When a check fails, the work goes back for correction within a set limit on attempts.
Related discussion on X ↗

Transparent costs

Understand spending at the level of each step, agent and model.

How it worksYou see what each step and each run costs, per agent and per model. The cost the agent reports and the estimated cost are always kept apart.
Related discussion on X ↗

A workflow that fits the job

Choose the right amount of structure for a small change or a complete delivery.

How it worksYou decide the path with blocks: long, or just three steps. Don't feel like drawing it? Describe it, and the Project Manager prepares it.
Related discussion on Reddit ↗

Clear completion criteria

Define what ready means through checks, reviews and your approval.

How it works“Done” has a clear rule you can see on the map: automatic checks first, then a second opinion, then your OK. Only then is a commit made.
Related discussion on X ↗

Your agents, coordinated

Bring Claude Code, Codex and other agents into one workspace with explicit settings.

How it worksAgent, model, keys and settings for each person live in one place, and each step uses the agent you picked, without silent swaps.
Related discussion on X ↗
An engine for any job

From Spec Kit without babysitting
to an engine for any job.

1

First: Spec Kit without babysitting

I started building rightcommit to run the Spec Kit process without watching an agent at every command. Specification, plan, tasks and implementation, with the right checks between each step.

2

Then: an engine for any job

The same building blocks turned out to be useful far beyond software development. Agents, tools, decisions, reviews and subworkflows can organize document analysis, research, reporting and operational processes, as well as the software development lifecycle.

3

Today: your workflow, beyond the template

Spec Kit is one starting point. By combining the available blocks and connecting the tools you need, you can build practically any workflow: define the work, decide who does it, verify the result and choose where a person needs to step in.

A visual workspace. An autonomous, scalable workflow engine.

Design workflows on the canvas. A powerful Rust engine orchestrates agents and tools, persists execution state, and enforces quality checks and controlled rework.

Run autonomously within your policies, pause for human decisions, and resume seamlessly. Scale the work with configurable concurrency, parallel task branches and reusable subworkflows. Execution queues and controls are built into the engine.

Persistent executionConfigurable autonomyParallel tasksReusable subworkflows

How much do the agents decide? You choose.

Self-configuring

Describe the job.
It builds your workflow.

A built-in assistant called Project Manager turns your goals into workflows. Describe what you need: it can create or refine the flow, check its structure, and propose changes you can apply, edit or undo. It can also follow the entire workflow, inspect artifacts and execution evidence, and provide targeted support.

Project Manager · Pi
Proposed draft
structure checkedApplyUndoSave & run

What you give it

  • Your goal, in plain words
  • A fresh snapshot of your current draft with every message
  • From a run: outputs, decisions, file versions and events
  • Your questions: “why did it stop?”

What you get back

  • A new workflow or changes to your draft
  • Checked proposals: if the structure doesn't hold, it tries to repair it in the same chat
  • Clear warnings about what's left to configure

What it never does

  • Never starts, approves or changes a run
  • No outside tools: it works in its own private space
  • Saved credentials are masked
Integrations

Automation chains
with the tools you already use.

Tickets, repos and docs join the map as blocks. Every write to the outside world stops and waits for your confirmation of the exact content.

GitHubissues · CI checks · draft PR
Azure DevOpswork items · draft PR
ServiceNowincidents · work notes
Jiravia Atlassian CLI
Confluencevia Atlassian CLI
Bitbucketvia Atlassian CLI
MCPtools from any server
A2Aagent to agent
Agents doing the workClaude CodeCodexPi

Jira + GitHub → a plan ready for review

ready-made workflow
Read the Jira ticket and the GitHub issue
Match ticket and code
Spec, plan, tasks, test strategy
Ready for your review

ServiceNow incident → fix → notes back

building blocks
Read the incident
Find the cause, fix it
Tests run by the server
Local commit
Work notes, after your OK

Issue → code → draft pull request

building blocks
Read the issue or work item
Plan and code
Tests run by the server
Local commit
You push the branch
Draft PR, after your OK

Release → change dossier

ready-made workflow
CI checks of the current commit
Read the release ticket
Change dossier
Human review

Started by another system

A2A
Another agent or another rightcommit
Starts your published workflow
Questions and answers in the same run
Returns the outputs you chose
connected servicecheckresultyou

Atlassian products are reached through Atlassian's official command-line tool (Teamwork Graph CLI): each command is confirmed with its exact arguments before it runs. MCP and A2A are open standards: MCP plugs tools into agents, A2A lets agents talk to each other. Service credentials stay on the server and never reach the agents. Pushing the branch stays with you.

MCP & A2A

Tools for your agents.
Workflows other agents can call.

MCP · Connect the tools

Give agents the tools they need from your MCP servers. Choose which tools each agent can use, scoped to the workflow step.

A2A · Make your workflows callable

Publish a workflow as an authenticated A2A agent. Other agents, including another rightcommit instance, can start it, follow its status and receive the outputs you choose. Your checks and approvals still apply. An A2A node can call other agents, too.

FinOps

Know what every agent
and every step costs.

FinOps simply means knowing what your AI agents cost you. In rightcommit it's a button next to Storage: open it and see where tokens and money went.

FinOps · illustrative example, not real figures
SummaryWorkflowAgentsModelsRunsManager
Reported costN/A
Estimated costN/A
Known total incl. estimatesN/A
Plan & tasksClaude Code
FixClaude Code · 2 attempts
Second opinionCodex · estimated
Test checkserver
  • Cost per step, run, workflow, agent and model
  • Input and output tokens, cache read and write, reasoning when the agent exposes it
  • Repeated attempts, fix loops, human waits and reviews
  • Cost reported by the agent and estimated cost always kept apart
  • Estimates use OpenRouter's public price list, with date and source shown
  • If a price is missing, the cost stays empty rather than showing a made-up zero
  • Charts by model, workflow, agent and step, plus a daily trend
  • The Project Manager has its own cost tab too

For consultants

You know what a client job cost in AI, step by step: a concrete basis for quotes and final reports.

For teams

Keep spending under control by seeing which workflows, agents and models cost the most and where repeated fixes increase the bill.

It's a reading of the usage the agents record, not an invoice. Infrastructure and outside-tool costs are not included.

Who it's for

On your own, with clients,
or across a company.

Developers

Like me: on your laptop, with your keys. A plan → code → tests → second-opinion loop that runs while you do something else.

  • Free to self-host
  • See what every step costs
  • Nothing gets published without you

Consultants

A repeatable method you bring from client to client, with proof of what was done.

  • Workflows you reuse and adapt per client
  • Decisions, versions and costs you can hand over
  • The license explicitly allows professional services for your clients

Enterprise teams

On your servers, with your tools: tickets, repos and docs flow into the workflows.

  • Multi-user: separate keys, workspaces and chats
  • ServiceNow, Jira, Confluence, GitHub, Azure DevOps
  • Every outside write waits for a confirmation
Free to self-host · Fair Source

Free to use.
Yours to build on.

Run rightcommit on your own machine or servers, for yourself, your team or client projects. Read the code, adapt it to your needs and share improvements under the Functional Source License (FSL-1.1-ALv2).

Make it your own

Use it at work, customize your workflows and build professional services for your clients. Internal use and professional services for clients using rightcommit under the license are explicitly permitted.

What it costs

No rightcommit license fee for permitted self-hosted use. You cover your infrastructure and any AI model usage with your own accounts.

How to contribute

Pull requests are welcome. On your first one, sign a CLA (contributor license agreement) through CLA Assistant with your GitHub account. You only need to do this once.

License text on fsl.software ↗ · CLA Assistant ↗ · Repository link coming soon

Get started

Running in 3 steps.

You need Docker and a key or login for at least one agent.

01

Start the container

From the rightcommit folder.

docker compose up -d --build
02

Create the first account

Open localhost:8080. The first user becomes admin; public sign-up closes after that.

open http://localhost:8080
03

Connect an agent, pick a workflow

In settings add your key or login and pick a model. Then press +, choose a ready-made workflow (or ask the Project Manager) and run.

FAQ

Questions & answers

How much does it cost?+

It's free to install and run yourself under FSL-1.1-ALv2. You only pay for the AI models you use, with your own keys or subscriptions.

Is it open source?+

rightcommit is Fair Source: the code is available to read, use, modify and share under FSL-1.1-ALv2. You can self-host it for work and provide professional services to clients within its terms. Competing commercial products and services are restricted by the license. Each version becomes fully open source under Apache 2.0 after two years.

Do I need to know Spec Kit or a special method?+

No. Start from one of the 11 ready-made workflows or describe what you need to the Project Manager. Spec Kit is included for those who want it, but it's just one path.

Which agents and models can I use?+

Claude Code, Codex and Pi, with an API key, login or gateway (Anthropic, Bedrock, Vertex, Foundry, OpenRouter and more). Each person picks their model and default agent.

Will it push, merge or deploy my code?+

No. Ready-made workflows end with a Git commit on your machine. Draft PRs on GitHub or Azure DevOps are created only after your confirmation, on a branch you've already pushed yourself.

Where do my code and keys live?+

On your machine or your servers. Keys are encrypted in the database (AES-256-GCM) and never sent back to the browser; credentials for connected services never reach the agents.

Why Rust? Is it heavy?+

The server is a single compiled Rust program that includes the web app and stores data in SQLite. It runs without additional database services. By default it listens only on localhost:8080. The container also ships the three agents' command-line tools and the Node.js runtime they need.

Try it

Let the agents work.
Keep the control.

Start with a ready-made workflow, connect your agent and follow every step through review.

Run it locally → Share your feedback