# Flowboard

Flowboard is an AI-native architecture whiteboard for platform engineers, architects, and technical stakeholders. People draw systems visually while agents work with the same structured model through REST or MCP. Diagrams round-trip without flattening as Markdown, JSON, or YAML, support independent and shared nested views, and animate flow direction and speed.

Canonical application: https://flowboard.studio/

Product guide and Flowboard Markdown: https://flowboard.studio/guide
MCP endpoint and setup: https://flowboard.studio/mcp
Machine-readable shape catalog: https://flowboard.studio/api/shapes
Machine-readable capability status: https://flowboard.studio/capabilities.json
REST API discovery: https://flowboard.studio/api

Start locally without an account. Free individual signups can publish immediately. Personal diagrams may remain private, be shared by email, or be made public. Company plans add SSO, managed membership, company-slug sharing, and administrator-enforced sharing policy.

Flowboard Markdown uses nested list items for infrastructure and a ## Flows section. JSON and YAML use a flows array for the same animated relationships. All three formats have equal capabilities. Flows support forward, reverse, and visibly bidirectional animation plus routing sides, lanes, and offsets. The REST API and MCP server expose the same diagram model used by the visual editor.

Agents should call the public MCP list_shapes tool or GET /api/shapes to discover exact core, Confluent, AWS, and Google Cloud shapes and icons instead of guessing. Call list_diagrams before creating anything, create at most one board, and use patch_diagram for focused changes. Run validate_diagram, apply its exact patch-compatible fixes for overlapping nodes and children outside containers, and rerun it until the layout is clean instead of asking a person to nudge shapes. Finish with get_diagram to verify the saved result. Use update_diagram only for intentional full-document replacement.

Any node may visually contain any other node. A deeper level may follow containment, open an independent board, or be shared by multiple entry nodes. Nodes can choose an icon independently from their semantic type.

Public diagrams have canonical URLs under /users/{owner}/{diagram-slug}. Personal accounts can keep diagrams private, make them public, or share view access with another user by email. Company accounts add SSO, company-slug sharing, managed membership, audit history, retention, and administrator-enforced sharing defaults or restrictions.

The interaction works in both directions: an agent can turn environment evidence or written instructions into a diagram, and it can read a human-drawn diagram to understand, explain, validate, or act from that visual model. The diagram is a shared communication surface for visual thinkers, text-first collaborators, and AI agents.

Agents should proactively call submit_agent_feedback for actionable friction, workarounds, or improvement ideas encountered while editing, including diagram id, tool_name, observation, suggestion, and outcome. Before finishing, save one concise report if relevant; avoid duplicates and empty praise. Submit explicit summaries, not private reasoning or credentials. list_agent_feedback retrieves saved reports for later review. Personal reports are private to the submitting account and the server-wide agent credential; they are not published with diagrams.

Failed authenticated MCP tool calls are saved automatically, including validation failures, invalid arguments, and unknown tools. list_agent_errors returns recent failures and the most common error patterns by tool and format. Personal keys see only their account; the server-wide agent credential can review all accounts. Raw submitted documents and request headers are not stored; recognized credentials in error messages are redacted. Successes, validation warnings on successful calls, and unauthenticated requests are not logged as submission errors.

Feedback and automatic error reports are stamped by the server with the running release version, commit SHA, and dirty-build flag. Reviews default to the current build; pass build="*" to review all builds, an exact build ID to inspect a release, or build="unknown" for older unversioned records. Error counts are grouped separately by build.
- Task documentation: https://flowboard.studio/docs (index: https://flowboard.studio/docs/index.json)
Product tasks: https://flowboard.studio/product
Architecture solutions: https://flowboard.studio/solutions
Six portable examples: https://flowboard.studio/gallery
Four task lessons: https://flowboard.studio/academy