Agents Machine

Team Workflow

How teams of 2–10 developers use Agents Machine for shared memory, coordinated agents, and team automation.

For development teams, Agents Machine becomes a shared AI brain — every team member benefits from collective knowledge, consistent code standards, and automated workflows.

Setting Up for Teams

Shared Memory = Shared Knowledge

When one developer stores an architecture decision, every team member's AI knows about it:

Store in memory: API authentication migrated from JWT to session tokens.
All new endpoints must use the requireAuth middleware from auth.plugin.ts.
Migration completed by @alex on 2025-03-01.
Category: architecture, Tags: auth, api, migration

Now when any team member asks their AI about authentication, it returns the latest decision — not outdated patterns.

Team Coding Standards

Store your team's standards once, enforce them everywhere:

Store in memory: Team coding standards:
1. All React components must be functional with TypeScript
2. API responses use ok()/fail() helpers from @shared/utils
3. Database queries go through repository layer, never direct Drizzle
4. All user input validated with Elysia t.Object() schemas
5. Errors use AppError classes from @core/errors
Category: coding-rules, Tags: team-standards, typescript, react

Project Isolation

Each project gets its own isolated memory, vault, and kanban. Team members working on different projects never see cross-contaminated context.

Daily Team Workflow

Automated Standup

Set up a pipeline that generates standup reports from actual task data:

Create a pipeline called "team-standup":
1. Trigger (schedule, cron: "0 9 * * 1-5")
2. Agent (manager) — "Generate standup report from kanban:
   - What was completed yesterday
   - What's in progress today
   - Any blockers or dependencies"
3. Skill (slack-notify) — post to #standup

PR Review Pipeline

Automate first-pass code reviews:

Create a pipeline called "pr-review":
1. Trigger (webhook, path: "/hooks/github-pr")
2. Memory Search — find relevant coding standards and architecture decisions
3. Agent (reviewer) — review against team standards + memory context
4. Agent (analyst) — analyze impact and regression risks
5. Merge — combine both reviews
6. HTTP Request — post review as GitHub PR comment

Knowledge Sharing

When a developer solves a tricky problem, they store the solution:

Store in memory: PostgreSQL connection pooling issue resolved.
Root cause: Drizzle default pool size (10) too low for our workload.
Fix: Set pool.max=25 in drizzle.config.ts.
Symptom: "too many clients" errors under load.
Category: service-context, Tags: postgres, drizzle, connection-pool, debugging

Next time anyone on the team hits a similar issue, the AI surfaces this solution automatically.

Team Vault

Shared Secrets, Individual Access

Store team API keys in the encrypted vault:

set_secret("GITHUB_TOKEN", "ghp_...", category="api_key")
set_secret("SLACK_WEBHOOK", "https://hooks.slack.com/...", category="webhook")
  • All team members can use secrets through skills (injected at runtime)
  • Secrets never appear in logs or AI responses
  • Full audit trail shows who accessed what and when

Kanban for Coordination

Sprint Planning with AI

Spawn the manager agent to decompose this feature into tasks:
"Add real-time collaboration to the document editor"
Create a board called "collab-feature" and add the tasks with priorities.

The manager agent:

  1. Queries memory for existing architecture
  2. Breaks the feature into concrete tasks
  3. Assigns priorities (P0–P3)
  4. Identifies dependencies between tasks

Task Handoffs

When a developer finishes a task, the AI updates memory automatically:

Mark task collab-03 as done.
Store in memory: Real-time cursor positions implemented using
Yjs CRDT with WebSocket transport. See src/collab/cursors.ts.
Category: service-context, Tags: collaboration, yjs, websocket

The next developer picking up a related task gets full context.

Scaling Patterns

New Team Member Onboarding

New developers get instant access to the entire team knowledge base:

Query memory for all architecture decisions.
Query memory for coding standards and anti-patterns.
Query memory for recent service context.

Instead of reading months of Confluence docs, the AI provides personalized, relevant context.

Retrospectives

Spawn the manager agent to run a sprint retrospective:
- Summarize completed tasks and their outcomes
- Identify patterns in blockers
- Suggest process improvements based on sprint data

Team Plan Benefits

FeatureFreeProTeam
Projects1UnlimitedUnlimited
Users115
Shared MemoryYes
Team VaultYes
Admin DashboardYes
Project IsolationYesYes

On this page