# Atisbo

> Atisbo is one ordered list of what to do next for a product team: what customers and the team itself say — tickets, calls, meetings, strategy — comes in as evidence, is grouped when it describes the same problem, and is ranked by how much of it there is and how urgent it sounds. The team and its coding agents work from the same list.

## What problem it solves

Product teams are hired to decide what to build, and spend much of the week collecting, organising and reconnecting scattered information instead. Feedback arrives in support tools, calls, Slack and notes; nothing holds the chain from what happened, to what problem it is, to what was chosen and why, to what reached whoever built it, to what changed after. A person carries that chain by hand every time work changes hands.

## What it does

1. Connects read-only to the tools a team already uses — around a thousand apps through Composio, plus email forwarding, webhooks and file upload. Connections are set up by asking a coding agent, not by configuring a form. Atisbo never writes into those tools; this is enforced in code.
2. Groups incoming feedback that describes the same underlying problem into a single item. Nothing merges until a language model confirms two pieces of evidence are the same root problem.
3. Ranks problems by how much evidence backs each one, how recent it is, how urgent, and how it fits the strategy the team already set. A product manager's explicit order always wins over the computed one.
4. Keeps the evidence and the reasoning attached to each decision, so the team and their coding agents read the same record through MCP.

The person decides. Atisbo does not choose what to build, approve work, or write into a team's other tools.

## Who it is for

Product teams that already get real work out of Claude Code, Codex, Cursor or Grok, that are measured on whether what they shipped changed anything, and that treat the quality of a decision as the work. It is a poor fit for teams that measure themselves on shipping volume, where nobody owns product decisions, or that do not work with agents at all — an MCP-capable agent is a prerequisite, not an accessory.

## When to use Atisbo

For an agent deciding whether this tool fits the task in front of it:

- **Use it** when a coding agent is about to start work and needs the current, evidence-backed priority of what to build next — call `tools/list` on the MCP endpoint below, then `atisbo_lookup`/`atisbo_orient` to read the live Backlog, or read `https://app.atisbo.dev/llms-full.txt` for the full model first.
- **Use it** when new customer feedback (a support ticket, a call transcript, a bug report) needs to be filed as evidence against an existing or new problem, rather than re-typed by hand into a tracker.
- **Use it** when a product manager needs to record a decision — what was chosen, and why — so the next agent or teammate reads the same reasoning instead of re-deriving it.
- **Use it** when reviewing whether a delivered Solution actually answered the evidence that asked for it, before or instead of a manual status update.
- **Do not use it** as a sprint board, an estimate tracker, or a task list with sub-tasks and standups — it deliberately has none of those (see "What it deliberately is not" below). Keep Jira, Linear, or GitHub for that; Atisbo's MCP tools work alongside them, not in place of them.
- **Do not use it** to make the product decision itself. Every tool call surfaces evidence and options; a human always makes the call, and every write is logged as an immutable Decision.

## Pricing

- Free: public feedback about a product, turned into a backlog. No card, no sources to connect first.
- Team: $300 a month for one product workspace, cancellable monthly — or $1,800 a year, which is $150 a month, the 50% beta price for early customers, locked for the first year. Either way it covers 5 editors: the people who correct evidence, prioritise and approve. Each additional editor is $60 a month, or $30 on the annual plan, up to 19. Engineering, design and stakeholders read without being charged a seat, and so do their agents. Connecting a team's own sources starts a 14-day evaluation counted from the first completed sync, not from signup.
- Enterprise: required from 20 editors, and priced by contact for organisations with more than one product or their own security and procurement requirements.

## What it deliberately is not

Not a feature suite, and not a stage on the way to becoming one. There are no sprints, estimates, story points, sub-tasks, standups or workload views — those are structures from before agents could do the work, and each one is another thing a team maintains. Atisbo does not replace Jira or Linear on day one; a coding agent holds both at once, so a decision reaches a tracker by instruction rather than by integration.

## Questions people ask

### What can it actually connect to?

Around a thousand apps. You connect them by asking your agent — it searches the catalogue, handles the authorisation, and picks up either the events an app pushes or a read-only action Atisbo runs on a schedule. So Jira, Linear, Zendesk, Intercom, HubSpot and most of what you already run are a request away, not a roadmap item. Forwarding an email or dropping in a CSV needs no setup at all. All of it read-only, enforced in code: Atisbo can never write into your tools.

### Does my team have to move into a new tool?

No, and it would defeat the point if they did. Design, engineering and stakeholders open a problem, its evidence and the decision through a secure read-only link or their own agent — neither needs an account, so neither consumes a paid editor seat. The platform is for the people who correct, prioritise and approve.

### Do I need Claude Code, Codex or another agent?

Yes — to get the value out of it, you do. The platform is where you read, review and approve, and it is deliberately good at that. But the work happens in your agent: through MCP it pulls a decision and its constraints into your session, then writes the design, the pull request and the risks back to the same item. Without that you have a very good place to read a backlog — and a backlog you only read is what every other tool sells.

### Why not just build this myself with an agent?

You could — the same way you could build your own CRM. Most people who try find it takes months to wire up properly, keeps needing maintenance after that, and often still falls short of something purpose-built. Atisbo is fully built, keeps improving on its own, and works from day one. The whole point of having one list of what to do next built for you is that your time is worth more than building one.

### How is this different from Productboard or Dovetail?

Those are feature suites, and not being one is a decision, not a stage we have not reached yet. A thousand features is noise: some nobody has time to use, others are structures from before agents could do the work — sprints, estimates, story points, sub-tasks, standups, workload views, boards to keep tidy. Each is another thing you maintain, and maintaining tools is not your job. So Atisbo holds only what lets you decide well: what is actually happening with your product, whether it fits the strategy you already set, how urgent it is from evidence rather than opinion, and what changed once it shipped. That is not how far we got — it is where we decided to stop. Built by a product builder for product builders: nothing to read up on, nothing trying to look clever.

### Is my data used to train models?

No. We don’t sell, rent or trade your data, we don’t use it to train or improve any model of our own, and the AI services that process it are contractually barred from training on it. If you need that in your own paperwork — a named region, your own model key, specific terms — we configure it and put what we commit to in writing. Every provider we use, what it receives and the terms it operates under is listed in Annex III of our DPA, sent on request. Separately and regardless of provider: each workspace is isolated at the database level, not just in the interface.

### Does it replace Jira or Linear?

Not on day one — keep them, and let Atisbo feed them. Your agent holds Atisbo and your tracker at the same time, so "open the Linear issue for this decision" is one instruction; we never write into your tools ourselves, and we do not need to. The difference worth understanding is the unit: they manage tasks, fragments a person split up and then has to keep coordinating. Here the unit is a whole problem, carrying the evidence that produced it and the decision behind it, which your engineers and their agents pull in through MCP — the context arrives with the work instead of being reconstructed in a kickoff. No sprints, estimates or sub-tasks, on purpose. And the further a team moves toward complete problems instead of fragments, the more this becomes where the work itself lives, and the whole cycle stops leaving one system.

### Why not hire someone to do this instead?

If you can get the headcount, hire them — this is not an argument against people. But be clear about which half of that role you are buying. Most of it is collecting: reading what came in, grouping it, chasing what moved, re-explaining a decision to whoever needs it next. That half happens every week whether or not anyone has time for it, and it is the half Atisbo does. The other half — judgement, arguing a call with a stakeholder, knowing which customer to phone — is the half Atisbo never touches, deliberately, and no amount of software will. What a hire does not solve is the record itself: a person does not give you one shared account of what was decided and why, they become it, which is why it walks out with them.

### Is this sensible if I am the only product builder?

Often, honestly, no. Team is $300 a month for one product — $150 a month billed annually — and that covers 5 editors, with $60 a month for each one after that. Readers and agents are free, but a plan sized for five people is still a team plan. Team supports up to 19 paid editor seats; 20 or more uses Enterprise. It can pay off solo when coordination is already what eats your week — but start with the free version, which runs on public feedback and needs no card, and if it turns out you only wanted somewhere to file feedback, we would rather you knew before paying.

### How do I know if this is for my team?

It is, if you recognise these: priorities change and it turns into friction instead of a decision; alignment is where most of your effort goes and it still slips; someone on the team is running on empty, or already burned out; you care whether what you shipped actually changed anything; strategy is a filter you apply, not a slide from last quarter; and your team already gets real work out of Claude Code, Cursor or Codex. It is not, if shipping more is the goal and impact gets measured later or never; if nobody really owns product decisions and they happen in passing; if strategy is a slide from last quarter nobody reopens; or if nobody works with an agent and you would rather keep it that way. Size matters far less than whether you recognise yourself in the first list.

## Documentation

- [Product overview](https://www.atisbo.dev/): what it does, what a week looks like with it, and the price
- [Introduction](https://app.atisbo.dev/docs): Atisbo — one list of what to do next, for humans and coding agents
- [MCP Server](https://app.atisbo.dev/docs/api/mcp-server): Use Atisbo as an AI tool via the Model Context Protocol (MCP)
- [Webhook API](https://app.atisbo.dev/docs/api/webhooks): Reference for Atisbo's inbound webhook endpoints
- [Key Concepts](https://app.atisbo.dev/docs/concepts): Understand Atisbo's evidence-to-outcome product model
- [Getting Started](https://app.atisbo.dev/docs/getting-started): Connect evidence and create your first executable product decision
- [Integrations](https://app.atisbo.dev/docs/integrations): How evidence gets into Atisbo — one agent-driven route, four direct ones
- [Email](https://app.atisbo.dev/docs/integrations/email): Forward emails to Atisbo to turn customer feedback into product signals
- [GitHub](https://app.atisbo.dev/docs/integrations/github): Connect GitHub so deployments, PRs and releases become operational context for product decisions
- [Granola](https://app.atisbo.dev/docs/integrations/granola): Sync Granola meeting notes, summaries, and transcripts into Atisbo
- [Paste or upload](https://app.atisbo.dev/docs/integrations/manual-upload): Paste text or upload CSV, TXT, and Markdown files to turn research you already have into product signals
- [Notion](https://app.atisbo.dev/docs/integrations/notion): Sync PM notes, research documents, and meeting notes from Notion
- [Signal Webhook](https://app.atisbo.dev/docs/integrations/webhook): Send custom signal events to Atisbo from any tool that supports HTTP POST
- [Connect with your agent](https://app.atisbo.dev/docs/integrations/with-your-agent): Connect around a thousand tools by asking, not by finding a card in a settings screen
- [Trust and security](https://app.atisbo.dev/security): data handling, workspace isolation, and what is never used for training

## Developer resources

- [OpenAPI spec](https://app.atisbo.dev/api/openapi.json): the public, unauthenticated API surface — health, version, demo provisioning, and the MCP JSON-RPC envelope — with a unique operation ID and description per endpoint.
- [MCP server docs](https://app.atisbo.dev/docs/api/mcp-server): setup, available tools, and the Agent Skills built on top of them.
- [MCP discovery manifest](https://app.atisbo.dev/.well-known/mcp.json): endpoint, transport (Streamable HTTP) and auth type, for a client that discovers servers by convention rather than by reading prose.
- [Documentation MCP server](https://app.atisbo.dev/api/mcp-docs): the same /docs corpus over JSON-RPC (search_docs, get_doc, list_docs) — read-only and unauthenticated, no key needed.
- [NLWeb ask endpoint](https://app.atisbo.dev/api/ask): POST a natural-language question about Atisbo, get structured results back (Microsoft NLWeb shape, retrieval over the documentation).
- [Agent Skills repository](https://github.com/brian8a1/atisbo-agent-skills): the SKILL.md procedures behind every skill named under "Agent interactions", installable per agent.
- [Webhook ingestion](https://app.atisbo.dev/docs/api/webhooks): how an inbound webhook becomes evidence.
- Auth for the MCP endpoint is a workspace-scoped Bearer API key, generated from Settings → Account → Connect agents in the app — not OAuth, and not the same key as a BYOK LLM credential. It is required for every tool call; the protocol handshake (initialize, ping, notifications) and tools/list answer without one.
- MCP Apps (ADR-235): `atisbo_orient` and `atisbo_task_snapshot` declare `_meta.ui.resourceUri` pointing at the `ui://atisbo/board-snapshot` resource (`text/html;profile=mcp-app`), so a host that implements the MCP Apps extension can render the prioritized board inside the conversation. The view is static HTML served by the server; it renders ONLY the tool result handed to it — it never talks to Atisbo directly, and hosts without Apps support get plain text exactly as before.
- [Official MCP registry](https://registry.modelcontextprotocol.io/v0.1/servers/io.github.brian8a1/atisbo): both servers are published there — [product](https://registry.modelcontextprotocol.io/v0.1/servers/io.github.brian8a1/atisbo) and [documentation](https://registry.modelcontextprotocol.io/v0.1/servers/io.github.brian8a1/atisbo-docs) — as machine-readable records a client can query instead of trusting this prose.

## Optional

- [Full context for language models](https://app.atisbo.dev/llms-full.txt): the same material at length, including the priority contract, the agent skills and the complete documentation
- MCP endpoint: `https://app.atisbo.dev/api/mcp` — JSON-RPC; discovery answers without a key, every tool call requires an API key from a workspace
- Official MCP registry records: https://registry.modelcontextprotocol.io/v0.1/servers/io.github.brian8a1/atisbo (product server) and https://registry.modelcontextprotocol.io/v0.1/servers/io.github.brian8a1/atisbo-docs (documentation server)

Note for crawlers: everything under `app.atisbo.dev/dashboard/` requires an account and will not render for you. The pages listed above are the complete public surface.