Using as a Knowledge Base

Last updated February 24, 2026

AutoMD boards are just markdown files. That means they can hold anything — not just tasks. Here’s how to turn AutoMD into a lightweight knowledge base your whole team (and AI agents) can query.

Why this works

Most knowledge bases rot because they’re separate from where work happens. With AutoMD:

Knowledge lives next to tasks — decisions and context sit in the same boards people check daily
AI agents can query it — ask an agent to search across all your .md files for decisions, context, or prior art
It’s just git — your knowledge base is version-controlled, diffable, and backed up automatically

Decision log

Track important decisions so you never have the “why did we do this?” conversation twice.

# Decisions

- [x] Use PostgreSQL over MySQL @team #architecture priority:high
  > Decided 2026-02-10. Postgres has better JSON support for our
  > flexible metadata schema. MySQL was considered but lacked
  > recursive CTEs we need for nested boards.

- [x] Ship MCP v2 protocol first @alice #api priority:high
  > Decided 2026-02-15. V1 clients can upgrade incrementally.
  > Breaking changes documented in migration-guide.md.

# Pending Decisions

- [ ] Choose hosting provider for cloud version @bob #infra priority:high due:friday
  > Options: Fly.io (simpler), AWS ECS (more control), Railway (fastest).
  > Need to benchmark cold start times before deciding.

Now any team member (or AI agent) can search for past decisions:

Search all my boards for decisions about "hosting" or "infrastructure".
What options were considered and what was decided?

Meeting notes

Keep meeting notes as boards — action items become trackable tasks automatically:

# 2026-02-24 Sprint Planning

- [x] Review sprint-11 velocity @team #meeting
  > 34 story points completed, 8 carried over.
  > Biggest blocker was auth refactor taking longer than estimated.

# Action Items

- [ ] Break down auth refactor into smaller tasks @charlie #backend priority:high due:tuesday
- [ ] Schedule design review for new dashboard @alice #design est:1h due:wednesday
- [ ] Update CI pipeline for parallel test runs @bob #devops est:3h

# Notes

General context here. Links, references, screenshots — whatever
helps the team remember what was discussed.

Ask your agent to take notes

During a meeting, tell the agent: “Create a meeting notes board for today’s sprint planning. I’ll dictate the action items.” Then just list tasks naturally — the agent will format them with proper metadata.

Project briefs

Start every project with a brief that lives alongside the task board:

# Project Brief

- [x] Define project scope @alice #planning
  > **Goal**: Ship user search with filters by March 15.
  > **Non-goals**: Advanced analytics, saved searches (v2).
  > **Success metric**: 80% of users find what they need in <3 clicks.

# Technical Approach

- [x] Choose search backend @charlie #architecture
  > Using Elasticsearch. Considered Meilisearch but need
  > geospatial queries for location-based filtering.

# Open Questions

- [ ] How do we handle search indexing for large accounts? @charlie #backend priority:medium
- [ ] Do we need real-time search suggestions? @alice #product priority:low

Architecture Decision Records (ADRs)

Use a dedicated board for ADRs — one task per decision with context in the body:

# Accepted

- [x] ADR-001: Use event sourcing for audit log @team #architecture
  > **Context**: Need immutable audit trail for compliance.
  > **Decision**: Event sourcing with append-only log.
  > **Consequences**: More storage, but perfect auditability.

- [x] ADR-002: Markdown-first data model @team #architecture
  > **Context**: Evaluated JSON, YAML, SQLite, and Markdown.
  > **Decision**: Markdown. Human-readable, git-friendly, universal.
  > **Consequences**: Parsing overhead, but worth the transparency.

# Proposed

- [ ] ADR-003: WebSocket vs SSE for real-time updates @bob #architecture priority:high

Querying your knowledge base

This is where the AI agent integration pays off. Your agent can search across all boards:

Search all my markdown files for anything related to "authentication".
Give me a summary of decisions made, tasks in progress, and open questions.

What architectural decisions have we made so far? List them with
their status and a one-line summary.

Find all open questions across all my boards. Group them by project.

The more context you write, the smarter your agent becomes

Agents can only find what exists. When you make a decision, write a sentence about why. When you close a task, add a note about the outcome. This context compounds — in a month, your agent can answer questions about your project that would take a human an hour to dig up.