Halyard Docs
Halyard is one knowledge layer with two readers. AI agents and the humans they work with share the same memory: agents read and write it over the Model Context Protocol (MCP); humans read and write it in Slack and the Halyard app. Nobody maintains a separate wiki — the layer fills itself as work happens.
That layer runs on a single loop:
- Search — an agent checks what the org already knows before doing anything else.
- Ask — on a genuine miss, the question routes to the right human expert in Slack.
- Capture — the answer (or the work output) becomes a knowledge entry the next agent can find.
Agents connect to one canonical endpoint:
https://mcp.usehalyard.aiWhere to go next
Section titled “Where to go next”Pick the path that matches what you’re here to do.
You’re setting up Halyard for a team
Section titled “You’re setting up Halyard for a team”You own the workspace: connect Slack, add experts, and get the org ready before any agent connects. Start with Set up your workspace.
You’re connecting an agent
Section titled “You’re connecting an agent”You’re wiring an AI client (Claude Code, Codex, Cursor, OpenCode) to Halyard. Get one client running the loop in under five minutes with the Quickstart, then see Connect an agent for client-specific setup.
Understand the model
Section titled “Understand the model”Once you’re connected, How Halyard works explains the loop in depth — including why an agent’s question often gets answered straight from the knowledge base without ever interrupting a human.