The Stable Baseline Agent.

Expand

What it can do

The agent covers the full span of Stable Baseline — documents, diagrams, improvements, plans, navigation. You don't pick a tool; you just ask for the outcome.

• Documents & content — search across your documentation; create, edit, or organise documents and folders.
• Diagrams — create diagrams in Mermaid, PlantUML, BPMN, D2, and 35+ other types. Edit existing diagrams — whole or part — and find architecture icons.
• Improvements & actions — track action items and improvements, add evidence and comments, manage categories.
• Plans & tasks — create and manage project plans with phases, add tasks, set priorities and dates, track progress.
• Navigation & more — jump to documents, projects, or settings, fetch external web content, answer open questions.

It understands what you're looking at

The agent watches the view you're on and narrows its scope accordingly — no need to paste workspace IDs or project slugs. Opening a document and asking “update the auth section to mention MFA” automatically targets this document. On a project dashboard, “summarise open improvements” searches this project. At the workspace level, the scope widens automatically.

• Document view → agent's scope is the current document and its linked items.
• Project view → scope expands to the project's docs, plans, and improvements.
• Workspace view → scope spans every project you can see.

Built-in todo management

Complex requests get broken into todos you can watch in real time. The agent marks items done as it makes progress; you can check items off yourself, edit them, or delete the whole list if priorities shift mid-task. Your todos, your workflow — the agent just keeps you informed.

Automates plans, tasks, and improvements

The agent doesn't just write docs — it drives the work. Paste a meeting transcript, a Loom summary, or a raw customer email into the panel and it can:

• Reshape a plan — adjust phase dates, re-prioritise, add or cancel tasks based on what was decided.
• Assign work — drop the right people onto the right tasks, with an acceptance criteria pulled straight from the discussion.
• Draft comments & activity — leave comments on tasks, plans, and improvements so the thread lives alongside the work, not in your inbox.
• Log improvements — every “we should probably…” turns into a tracked improvement with evidence linked back to the doc or ticket that sparked it.

Frontier models, handled for you

You don't bring an API key and you don't pick a model. We route every call through zero-data-retention providers and pick frontier-class models on your behalf to ensure the best answer for the task — reasoning-heavy work gets a reasoning model, fast look-ups get a fast one.

What costs credits — and what doesn't

Chatting with the agent is free — no credits deducted, no throttling, no “out of message” wall. Credits are only spent on the heavier work:

• Web fetches — when the agent pulls in external pages or APIs to ground an answer.
• File uploads & parsing — attaching a PDF, spreadsheet, or image for the agent to read.
• Deep research — multi-step investigations that fan out across external sources.

Each credited call is shown inline in the chat the moment it happens, so your balance is always visible and you can stop or adjust scope before anything heavy runs.

How to open the agent

Click the Stable Baseline Agent button at the top right of any view. The panel slides in from the right and stays pinned while you work — close it any time from the × icon.

On a document, the agent sits alongside the editor. On a project or workspace view, the panel opens on the right with no loss of your current context. The panel carries its own history per scope, so switching between a document and the project doesn't reset your conversation.

Starter prompts

From a blank panel, things to try:

• “What's in this doc? Summarise for someone new to the project.”
• “Draft a runbook for a Postgres failover. Use a BPMN swimlane for the on-call / SRE / comms handoff.”
• “Show me all open improvements with priority ‘critical’ and what's blocking them.”
• “Create a plan for Q3 with three phases: discovery, build, launch. Pre-populate tasks from improvement IMP-14.”
• “Find every doc that mentions the deprecated auth flag and suggest edits.”
• “Take me to the project settings page.”

Vs. bringing your own agent over MCP

The built-in agent is the fastest way to get value — one click and you're working. It's ideal for most day-to-day document and plan work.

If you'd rather drive Stable Baseline from your own IDE-based agent (Claude Code, Cursor, VS Code, Windsurf, …), do that through MCP instead. Same features, same tools, different entry point.

Which should you use?

• Built-in agent — for document writing, planning, answering questions across your workspace, quick diagram generation.
• IDE agent over MCP — for editing code and docs side-by-side, keeping documentation in sync with commits, running multi-step tasks against your repo.
• Many teams use both — the built-in agent for workspace work, and MCP agents for code-aligned work. They share one knowledge graph, so edits from either side flow through immediately.