Where people and agents share one understanding
Your team and its coding agents work from the same plan, side by side. Piyaz keeps handing them ready work until the project is delivered. You stay in charge of what ships, and the loop handles the rest.
Agents can do the work but they skip the discipline
The context is lost
The work collides
You already have a tracker
An agent working alone skips every step you'd insist on from a person: the brief, the agreed definition of done, the plan, the review, the sign-off. It skips them because there's nowhere to put them, and nothing anywhere that says what the work depends on.
Nobody has to explain the work twice
Each piece of work carries its own brief. It says what needs to happen and what finished looks like, and it keeps the decisions and the history behind it. Whoever picks it up, a person or an agent, starts with the full picture, so time goes into the work instead of catching up.
Snap the camera on resize
A phone collapsing its address bar changes the viewport mid-scroll, and the motion setting can flip while the page is open. Both change what the camera should be framing, and animating toward the correction reads as a glitch. The loop re-arms a one-shot snap and lands the new framing in a single paint.
- A resize lands the new framing in one paint
- Flipping the motion setting reframes without a reload
- Ordinary scrolling still eases into frame
- Nothing else in the render loop restarts with it
- Snap instead of animate, a camera correcting itself in view reads as a bug
- Arm the snap inside the loop restart so a new caller cannot forget to
Everyone sees the same project
Describe an idea, or point Piyaz at work you have already started, and it breaks it into clear pieces and works out what depends on what. Handovers happen through the project itself, so nobody has to sit in the middle relaying messages. At any moment you can see what is finished, what is ready to start, and what is waiting on something else. And the whole team works on it at once, people and agents side by side.
Whoever starts a piece of work claims it first, so a person and an agent never land on the same task.
Only pieces that do not depend on each other run at the same time.
Each one works in its own space, so nobody overwrites anybody's changes.
Projects belong to your team. Assign any task to a person or an agent, hand work between them mid-project, and see everything on your own plate in one list.
“A change that does not reach everyone did not happen”
One place for what your team knows
Notes hold the way your team works, and the reason behind decisions nobody will remember in a month. Mark one as always on and it lands inside the brief for every task it covers. The others stay just a title and a line. An agent opens those when the work needs them.
Agents can add to it freely, and anyone can search what they add straight away. But it only starts shaping other people’s work once a person says so. An agent can ask for that. It cannot decide it.
Work keeps moving until it is done
Plan it
Describe what you want. Piyaz asks the questions a good lead would ask, pushes back on the weak parts, and turns the answer into clear pieces of work.
Pick it up
Ask what to work on next and Piyaz names the piece that matters most, hands over the full background, and records what happened when it is finished.
Review it
Nothing we ship promotes its own work to done. A second agent reviews it first and comes back with approve, changes needed, or stop, pointing at exactly what it found.
Or hand the loop to Composer
In Claude Code, Composer runs the cycle on its own: research, plan, build, review, fix, then the next piece. It keeps going until the work is done, and you stay the one who approves.
You approve what ships not every keystroke
Agents do the work and write down what they did. You stay above it, approving what goes out, instead of being pulled into every small decision along the way. Work through the loop piece by piece, or let Composer run ahead and bring you the finished result.
The agents we ship stop at review and hand back a verdict citing the files they touched. Marking it done is yours.
You bring the intelligence
Piyaz runs no models, never asks for an API key, and does not charge for intelligence. The agents your team already uses do the thinking, in whichever harness you prefer: Claude Code, Codex, Cursor, or Antigravity.
What Piyaz adds is the shared ground they deliver on: one plan, your team’s knowledge, and a person approving what ships. We build Piyaz using Piyaz, and the project on this page is our own.
Nothing to set up
Piyaz runs at app.piyaz.ai. Create an account, connect your agent once, and you are working. There is no server to run and no upgrade to chase.
Simple while in beta
One plan while we are in beta, at no cost today. Paid plans will be announced on the pricing page.
Open, so you can check
The code is public under AGPL-3.0. Anyone can read exactly how your project information is stored and used, which matters more for a tool that holds what your team knows.
Prefer to run it on your own infrastructure? You can. Piyaz is self-hostable, and it works the same either way. The code is on GitHub.
The things people ask first
- What is Piyaz?
- Piyaz is a shared workspace for teams that work with coding agents. It keeps people and agents on the same plan, so everyone can see what is done, what is next, and why each decision was made. And it keeps the loop moving, handing agents ready work until the project is delivered.
- How much does it cost?
- While Piyaz is in beta there is one plan and it costs nothing. Paid plans will be announced on the pricing page.
- Which coding agents does it work with?
- Claude Code, Codex, Cursor, and Antigravity. Piyaz plugs into the agent you already use. You connect it once and it reads and writes the same project your team does. The connection is an MCP server, so anything that speaks MCP can reach the same project.
- Do I need an API key?
- No. Piyaz runs no models of its own and never asks for a key. Your agents bring the intelligence, and piyaz.ai is where they meet your team and each other.
- Do you charge for the intelligence?
- No. Intelligence is not something Piyaz charges for. Your agents bring their own, in whichever harness your team prefers: Claude Code, Codex, Cursor, or Antigravity.
- Do the agents decide what ships?
- No. You set the direction. An agent picks up work when it is ready and stops before marking its own work finished. What ships is your call, and the loop stays under your control.
- Can Piyaz run the whole loop by itself?
- In Claude Code, yes. Composer can take one task, a feature, or a whole project and run the loop end to end: it researches and plans each piece, builds it, has it reviewed, and fixes what the review finds, then moves on to the next one. A person still decides when to stop.
- Do I need to install or host anything?
- No. Piyaz runs at app.piyaz.ai. Create an account, connect your agent once, and your projects are reachable from any machine. Self-hosting is also possible, since Piyaz is open source, but nothing about the hosted version requires it.
- What happens when a conversation ends?
- Nothing is lost. Decisions, plans, and what earlier work actually did stay recorded against each piece of work, alongside your team's shared knowledge. The next person or agent starts informed instead of from nothing.
Put your team and your agents on the same page
Piyaz is in beta and open for signup. Create an account, connect the agent you already use, and pick up your project from any machine.