Agent Workflows

How to integrate Buildd Memory effectively into your AI agent's workflow.

The Context → Work → Save Pattern

The recommended workflow for agents using Memory:

1. Load Context at Session Start

At the beginning of every session, load relevant memories:

memory({ action: "context", params: { project: "my-project" } })

This returns all memories for the project formatted as markdown. Inject this into the agent's context so it starts with team knowledge.

2. Search When Needed

During work, search for specific memories when encountering unfamiliar code or making decisions:

memory({ action: "search", params: { query: "database transactions" } })

Use search to find:

Known gotchas before making changes
Architectural decisions that affect your approach
Patterns to follow for consistency

3. Save What You Learn

When you discover something important, save it immediately:

memory({
  action: "save",
  params: {
    type: "gotcha",
    title: "Neon HTTP driver doesn't support interactive transactions",
    content: "Use atomic UPDATE...WHERE with .returning() for optimistic locking instead of db.transaction().",
    project: "buildd",
    tags: ["database", "neon", "drizzle"],
    files: ["packages/core/db/schema.ts"]
  }
})

What to Save

Do save:

Gotchas that cost you time
Architectural decisions and their rationale
Patterns that are specific to this codebase
Non-obvious behavior you discovered
Key file paths and their purposes

Don't save:

Obvious or well-documented behavior
Temporary workarounds
Session-specific context (current task, in-progress state)
Information that changes frequently

Using Tags and Files

Files

Associate memories with specific files so agents working on those files get relevant context:

files: ["src/lib/auth.ts"] — Memory about the auth module
files: ["packages/core/db/schema.ts"] — Database schema gotchas

Multi-Project Setup

Use the project field to organize memories by project:

// Save with project scope
memory({
  action: "save",
  params: { type: "pattern", title: "...", content: "...", project: "buildd" }
})

// Load only that project's memories
memory({ action: "context", params: { project: "buildd" } })

Without a project filter, context returns all team memories.

Team Coordination

Memory works best when all agents on a team use the same API key. This means:

Agent A discovers a gotcha → saves it
Agent B starts a new session → loads context → sees the gotcha
Agent B avoids the same mistake

The more consistently your agents save useful knowledge, the smarter the whole team becomes.

Agent Workflows

On this page