Ask for it.
Watch it appear.
A project tracker where the UI isn't fixed. Describe the dashboard you need — Clay builds it against your real data, in seconds.
One request in, one working view out.
These aren't screenshots of a designer's mockup — they're what a session in Clay actually produces.
01 — ASK
Type a sentence, get a dashboard.
The card on the right is a view the agent produced from the single request at its top. It picked the chart, bound it to an approved query over the Website Relaunch project's real tasks, and rendered the result — no chart picker, no field mapping, no configuration screen.
The live view badge means exactly that: when a task changes status, the bars change with it.
“Show tasks by status as a bar chart.”
One sentence in chat. The agent assembled this view and wired it to the Website Relaunch project.
02 — REFINE
Keep talking until it's right.
This card is that same view's history. Each follow-up request — a bigger chart, colors by status — became a new numbered version instead of an overwrite, so you can step back to any point or branch off in a new direction.
The unchecked line isn't a task — it's simply the next request that hasn't been asked yet.
Refined in conversation.
Every request becomes a new version. Nothing is overwritten.
- “Make the chart bigger” — v2
- “Color the bars by status” — v3
- “Add a done-this-week tile”
Not a form builder.
Not another dashboard.
Just a tracker for projects and tasks — with an agent that assembles whatever interface you ask it for, out of parts that were checked before it got to use them.
01
Ask. Watch. Ship.
No drag-and-drop builder, no widget library, no settings maze. Type a sentence and the view exists.
02
Real data. Real team.
Every view queries the same live database that holds your projects and tasks — shared with whoever you invite.
03
Yours to rewind.
Each refinement is a new version. Step back through every iteration a view has ever had.
Small on purpose. Live by default.
Clay keeps the core loop simple: track your work, ask for the view you need, and keep every version on your side of the line.
Try It YourselfAsk, Don't Configure
Describe a dashboard in plain language and it exists seconds later — no chart picker, no field mapping, no settings maze.
Refine in Conversation
“Make it bigger.” “Group by assignee.” The agent remembers the thread, so follow-ups build on each other.
Versioned by Default
Every change is a new version, never an overwrite. Roll back to any point, with a real diff of what moved.
Guardrailed by Construction
The agent never writes code or SQL. It picks validated widgets and binds them to approved queries — so a generated view is safe the moment it appears.
Built for a Team
Invite people into a shared workspace, assign work, comment on tasks. Every view, chart and audit entry is scoped to it.
Connect Your Assistant
Clay speaks MCP. Point Claude Desktop at it and ask about your projects from anywhere — read-only, one workspace, revocable.
Run it, then ask.
A Docker Postgres and one dev command get a private Clay workspace running on your machine in minutes.
View the RepositoryThree commands to start your Clay server
docker compose up -d
npm install
npm run dev
Where Clay fits.
Use it wherever the view you need keeps changing faster than anyone wants to configure it.
Choose your next step.
Create a workspace, read the code, or run Clay locally and see how the agent is built.
Ask for your first view
Create an account, add a project, and describe the dashboard you want.
SourceRead the codeThe full app — agent loop, tools, and UI — is on GitHub.DeployRun it locallyPostgres via Docker, then one command for the dev server.TrustSee the guardrailsRead-only agent tools, rate limits, and a full audit trail.Common questions.
What Clay is, what the agent can touch, and what happens to your views over time.
What does the agent actually build?
Real interfaces — charts, tables, boards and tiles — rendered inside Clay against your live project data. It does not write code. It picks from a validated vocabulary of widgets and binds each one to an approved query, which is why a generated view can be trusted the moment it appears rather than reviewed first.
Can the agent change or delete my data?
No. It works through a small set of vetted, read-only tools: describe the data model, run an approved query, propose a view. Writing isn't one of its abilities. A view can still contain controls that change things — a status dropdown on a table row — but those only ever run when you click them, under your own account, never the agent's.
Can I try it without signing up?
Yes, and it isn't a read-only tour. The demo hands you a real workspace pre-filled with six projects and a team — you can create tasks, drag dashboards around, import a spreadsheet and export a workbook. It's the same application a paying customer uses; it just gets deleted after 24 hours.
Can I get my existing work in?
Import a CSV or an Excel file. Columns are matched automatically where the header is recognisable, statuses and priorities are translated from however your spreadsheet words them, and a dry run tells you exactly what will be created — and which rows need attention — before anything is written.
Does it work with my team?
Invite people into a shared workspace and everything is scoped to it: assignees, comments, workload charts, the audit log. Owners keep the actions that reshape a workspace, like deleting a project.
What is Clay built with?
Next.js and React on the front, tRPC and Drizzle over Postgres on the back, Clerk for auth, and the Claude API driving the agent loop. Around 250 tests, including a suite that tries to make the agent misbehave and a graded set that checks the views it builds are actually good. The source is on GitHub.