Mataara

Product management for teams that ship with AI agents

Your agents ship. Mataara keeps the record.

Mataara is product management for teams that ship with AI agents. It keeps one published plan of record, versioned and never rewritten silently, plus the commitments and goals behind it. Claude, Grok and your other agents read it. Agent access keys are on Business and Enterprise. Agents propose; a person approves every stage move and every publish. Every action is attributed.

Register See how it works
Publish · acknowledge · audit Agents connect with access keys Agents never accept their own proposals Connect GitHub (PRs, CI) Roadmap
One plan of record
People and agents read the same published version. Draft → Published → Superseded. Never a second truth in a side channel.
Agents draft and propose, and never accept their own proposals. The Security page shows exactly what each agent key can do.
Agents read the plan and propose changes. A person accepts or rejects stage and feature-pack proposals. Nothing rewrites a published version silently.
Every agent action attributed
Each agent has its own user and a named access key scoped to the cards you allow. Every call carries both.

Built on the platform

  • Managed Documents
  • Publish and acknowledgment compliance
  • Hosted agent endpoint
  • Scoped agent access keys
  • Tenant-isolated by design

Why now

Agents changed who does the work. They didn’t change what you have to prove.

01

The plan lives in a chat

Each agent gets its instructions from whoever prompted it last. There’s no single version of what was agreed.

02

Nobody can say who did what

Most trackers record the person who clicked “confirm”, not the agent that made the change.

03

“Done” means “the agent said so”

Without criteria the work is checked against, review turns into reading diffs at midnight.

Mataara fixes all three with one record: a published plan, agents with their own identities, and acceptance criteria that decide “done”.

What your agents work from

Six documents that run a product.

An agent is only as good as the record it reads. Mataara makes that record real, published documents — versioned, acknowledged, and readable by your clients, your executives, your engineers and your agents alike. Not cards on a board that only the product team can see.

Feature spec

The narrative and the acceptance criteria for one unit of work. The contract an agent builds against and a reviewer checks against.

FEATURE_SPEC

Roadmap

Now, Next and Later as a published document with a version number — so an agent knows what is in scope this quarter, not just this prompt.

ROADMAP

Product vision

Product and domain briefs that say why, so every spec below them — and every agent working on them — inherits the intent.

PRODUCT_VISION · DOMAIN_BRIEF

Release notes

What shipped and what changed, published alongside the plan it came from — whether a person or an agent did the shipping.

RELEASE_NOTES · CHANGELOG

Feedback note

Client feedback triaged into the same document model and linked to the feature it affects, so agents see the why behind the change.

FEEDBACK_NOTE

Stakeholder pack

The progress pack for the board or the client. Published, acknowledged, exported to PDF on your template.

STAKEHOLDER_PACK

And the record around them: a commitments ledger and OKRs, linked to the cards they shape.

Every document type carries a format template that agents fetch before they write, so a spec drafted by Claude has the same headings as one drafted by your team.

Examples

A week with Mataara and your agents.

Real Mataara tools, real gates. Agents read the plan of record and propose, and the proposing agent never accepts its own work. Here is what that looks like across one feature.

Monday

Ask AI for priorities you can accept

Ask AI to suggest priorities: every score shows its reasoning and sources, and nothing re-ranks until you accept.

Tuesday

Build against the contract, not the chat

A Grok Bot minted for delivery pulls the same Feature pack: identity, breadcrumb, spec, acceptance criteria and current stage. It builds, opens the pull request, and proposes the move to Testing. A person accepts the stage move. Nothing moved on the plan until they did.

  • get_feature_pack
  • get_feature_sdlc
  • propose_feature_sdlc
Wednesday

Know what a change touches before you make it.

Before you move or change a card, Mataara shows the blast radius: what it affects. Record the call before you go ahead.

Thursday

Scope an agent to exactly two products

The owner sets up a review agent in one step — Mataara mints a bot user, a named access key and a ready-to-paste config. The allow-list names two Products. Everything else is denied. One-step setup won't mint an empty allow-list, and an agent key left empty is refused every call. Revoke the key and the agent is out, instantly.

  • Setup external agent
  • Allow-list scopes
  • Revoke key
Friday

Publish to the client. Account for the agents.

Release notes and the progress pack go out as published documents. The client acknowledges the version they read. Behind them, the access-key usage log shows which agent key made each call, and when. Full agent report-back into an activity ledger is on the roadmap.

  • RELEASE_NOTES
  • STAKEHOLDER_PACK
  • Roadmap activity ledger
One proposal, end to end
  1. C Claude — Specget_feature_pack feature: “Client portal sign-in”
  2. Mataarapack v4 · spec · 6 acceptance criteria · format template · stage: Ideation · attributed to key ····7f3a
  3. C Claude — Specpropose_feature_pack_change section: Acceptance criteria · +2 criteria
  4. Mataaraproposal pending — live document unchanged, awaiting review
  5. TW Tui W. · ProductAccept → 2 criteria added · proposed by Claude — Spec · accepted by Tui W.

Why Mataara

Governance for product decisions, at the speed agents ship.

01

One plan of record for people and agents

A published document your stakeholders acknowledge and your agents read — not a private board link that changes under them, and not a prompt that lives in one person’s chat history.

02

Agents draft and propose, and never accept their own proposals. The Security page shows exactly what each agent key can do.

Agents propose. A person accepts or rejects stage and feature-pack proposals. The proposing agent cannot accept its own work, and a stale proposal fails rather than overwriting.

03

Every agent action attributed

Each agent is a named user with its own access key. Every call carries both, and the usage log shows which agent key made each call. Ready for the review, the audit, or the incident call.

04

Built for review, audit and the client call

Version history, review and acknowledgment on every document. Fail-closed tenancy. Allow-list scopes where empty means denied. One brand on a shared platform.

How it works

Plan. Publish. Connect your agents. Review their work.

Four moves, one document model. Agents join the record; they do not replace it.

1

Plan on cards

Lay out Product, Domain, Sub Domain and Feature. Collapse the levels you don’t need yet. The hierarchy is the map, not the artefact.

2

Publish the record

Draft specs, roadmaps and packs as managed documents under the cards. AI drafts inside the document; a person reviews and publishes. Stakeholders acknowledge the version they read.

3

Connect your agents and your repo

Agents get a scoped key. Roadmap A GitHub App that links pull requests, CI and deploys to Features and proposes stage moves is on the roadmap. Mataara never reads your code.

4

Accept or reject stage and feature-pack proposals

Agents read the Feature pack and propose changes to the spec, the criteria or the delivery stage. A person accepts or rejects stage and feature-pack proposals. Agents never accept their own proposals.

  • PRODUCT Client Portal
    • DOMAIN Access & identity
      • SUB DOMAIN Self-service onboarding
        • FEATURE Client portal sign-in
          • FEATURE_SPEC Spec + 6 acceptance criteria
          • AGENT Claude — Spec · proposal pending
          • AGENT Grok Bot — Delivery · PR open

The model

Cards for the map. Documents for the record. Agents on the leash.

Four levels, no more

Product, Domain, Sub Domain, Feature. A small team may run one product and one domain; the model still holds, and empty levels stay out of the way.

A Feature is a spec plus acceptance criteria

Given / when / then, stored as data on the Feature — human-readable for your team, machine-readable for agents. Agent suggestions are held as proposals until a person accepts or rejects those feature-pack proposals. A key or bot cannot edit criteria directly.

An agent is scoped to the cards you name

Each agent has its own user and a named access key. The key carries an allow-list of Products, cards and document types. For agent keys, an empty allow-list means denied: one-step setup won't create one, and any agent key left empty is refused every call. Revoke it and the agent is gone.

Controls

Everything a plan of record needs when agents are on it.

Mataara is built on a shared platform. The governance features below are the same publish-and-acknowledge engine, now extended to agents.

Plan

AI prioritisation

AI suggests RICE, ICE or Value/Effort scores with cited reasoning. You accept, it never re-ranks on its own.

OKR tracker

Objectives and key results with check-ins, linked to the Features that move them.

Commitments ledger

Track what you promised a client or stakeholder, and whether it was kept.

Blast-radius checks

See what a change touches before you make it.

Publish, acknowledge, re-acknowledge

Set the audience, capture reading evidence, require periodic re-acknowledgment, and see who has read what in one compliance view. The version people acknowledged is the one agents read.

Versions you can defend

Draft → Published → Superseded → Retired. Compare versions. Export to PDF on your own template. Nothing an agent does rewrites a published version silently.

Sources stay attached

Upload research, contracts and designs into the library. Link them under the card as evidence for the record, for people and agents to cite.

Roadmap

Real-time co-authoring

People draft together, live. Comments, tracked changes and approval sit on the document, not in a chat thread — and accepted agent proposals land in the same room.

Control

Hosted agent endpoint

Claude, Grok Bot and any agent connect to your tenant’s hosted endpoint. Planning cards, documents, Feature packs and delivery state are tools — the same permission-filtered API your people use.

Scoped agent access keys

One bot user and one named key per agent. Allow-list the Products, cards and document types it may touch. For agent keys, an empty allow-list means denied: one-step setup won't create one, and any agent key left empty is refused every call. Revoke at any time. Active keys: Plus 30, Business 60.

Propose, then Accept or Reject

Agents propose changes to the spec, the acceptance criteria and the delivery stage. A person accepts or rejects stage and feature-pack proposals. Stale proposals fail. The proposer cannot accept its own work. Keys never publish. A person approves every stage move and every publish.

Feature pack for agents

One packed read: identity, breadcrumb, spec, format-template headings, acceptance criteria and delivery state. Versioned, and scoped to the agent’s allow-list — an agent never sees a card it was not granted.

Attribution on every call

Every call carries the agent’s user and its access key. Owners and admins see the usage log per key, so “which agent did that?” has an answer. Sign-ins, AI runs, impact checks and commitment changes are audit-logged too.

Tenant-isolated by design

Fail-closed tenancy and permission-filtered tools. Agents always need a key, with no anonymous access. The same isolation the rest of the platform runs on.

Roadmap

GitHub App connector

On the roadmap. It will install on the repos you choose so PRs, CI and deploys land on the Feature, and stage moves are proposed, never forced. Mataara never reads your code.

Roadmap

Agent activity ledger

Agents report runs, tests and deploys back into an activity ledger behind the documents, so the progress pack accounts for the bots as well as the people.

Roadmap

Per-tenant rate limits

Plan-sized limits per tenant and per key, so a runaway loop or a leaked key can’t starve your team.

Who it’s for

Teams that ship with agents and answer to someone.

Software teams

You run Claude Code, Cursor or Codex on real work. Keep every agent on the spec and know what each one shipped.

Mataara for software teams

Agencies and dev shops

Get the client to sign off the exact version of scope, then show them delivery against it, on your letterhead.

Mataara for agencies

Security and delivery leads

Scoped keys, proposals and attribution. Exactly what’s live today and what’s next.

How Mataara controls agents

Honest comparison

Why not a board?

Boards are good at ranking. Agent dashboards are good at showing sessions. Neither is the thing a client signs off, and neither tells an agent what the plan actually is. Mataara does the same product-management jobs, but the published, acknowledged document is the product — and the agents work from it. Not affiliated with any product named or implied.

Alternative Their shape Mataara
Prioritisation boards The board is the primary object. Detail lives in cards and comments; stakeholders get a link or a screenshot. The published document is the primary object. Cards are the map; the versioned, acknowledged artefact is the record — for people and agents.
Prioritisation A board drags cards. Mataara shows why, with cited reasoning a person accepts.
Agent coding tools and issue trackers Track agent sessions per issue. The agent works from the ticket and the chat; the plan lives somewhere else. Agents read the plan of record itself — spec, criteria, stage — and their proposals wait for a person. The record is the source, not the ticket.
Roadmap suites Board-heavy roadmaps and maker seats, with exports for everyone else. The roadmap is a document with a version number and an acknowledgment list — readable by anyone you publish it to, including your agents.
Docs and wikis Flexible pages and AI, without a formal publish, acknowledgment or version-of-record process. Anything with access can edit. Draft, review, publish, acknowledge, supersede — a governed lifecycle on every document. Agents propose; they do not edit.
Delivery dashboards Answer “what did the bots do?” for engineers. Clients still ask what shipped and what changed. Release notes and progress packs for clients first; agent activity attributed per key behind them.

No scorecards, no price comparisons. Talk to us if you want the detail.

Questions

Straight answers.

Which AI agents can connect to Mataara?

Claude and Grok Bot have a one-step setup, on Business and Enterprise, that mints the agent’s key and hands you a ready-to-paste config. Other agents — including coding agents and your own orchestrators — connect to the same hosted endpoint with the same scoped key.

Can an agent change the plan on its own?

Only by proposing, apart from ordinary card edits. Agents propose changes to the spec, the acceptance criteria or the delivery stage, and a person accepts or rejects those stage and feature-pack proposals. A key or bot cannot edit criteria or move stages directly, including starting delivery or clearing a stage that is already set. Ordinary card edits stay allowed. Staging and production moves also need a deploy pin. An agent can still file that pin itself through report_bot_activity. The pin doesn't move the stage; a person still makes the move. The Security page shows exactly what each agent key can do. A published version is never rewritten silently. Agents also can’t accept AI priority suggestions or record blast-radius outcomes. Those actions are person-only.

How do I control what an agent can see and do?

Each agent has its own user and a named access key. The key carries an allow-list of the Products, cards and document types it may touch. For agent keys, an empty allow-list means denied: one-step setup won't create one, and any agent key left empty is refused every call. Revoke the key and the agent is out. Every call is logged against the key.

Is Mataara another prioritisation board?

No. Mataara can suggest priorities with AI, but each suggestion shows its reasoning and a person accepts it. Accepting never moves a card for you. The product is the published, versioned, acknowledged record underneath.

Does AI publish anything on its own?

No. Only a person publishes a managed or card document. That covers everything you can publish in Mataara. Every generation is a logged run.

Do I have to replace Jira or Linear?

No. Keep your tracker for tickets. Mataara holds the spec and criteria your tickets are built against, and the record of what was delivered.

Is Mataara only for software?

Today, yes. Software has the deepest tooling: pull requests, test runs and deploy pins. Teams also use it for governance and strategy documents. We’ll add other kinds of work when customers pull us there.

What does it cost?

You pay per maker, the people who write the plan. Reviewers, clients and stakeholders are free on every plan. Agent access keys are included on Business and Enterprise today. See Pricing; plans are shown when you register. Bots are free within each plan’s cap: Team 10, Plus 25, Business 50, Enterprise by contract.

Can I sign in with GitHub?

Roadmap Sign-in with GitHub, Google or Microsoft is on the roadmap, not available today. The GitHub App that would link pull requests, CI and deploys is separate, and that is on the roadmap too. Neither reads your code.

Ready when you are

Put your agents on the plan of record today.

Register, choose the plan that fits your organisation, and invite your team. Connecting your first agent is on Business or Enterprise. Plans and pricing are shown when you register.

  • One published plan of record your people and agents read
  • Scoped, attributable agent access keys. A GitHub App for PRs, CI and deploys is on the roadmap. Roadmap
  • Stage and feature-pack proposals your team accepts or rejects — nothing lands silently
Register Sign in

Compare plans: Pricing.

Enterprise rollout, contracts or retention requirements? Talk to us.

The app is at app.mataara.app. Marketing origin mataara.app.