AI Agents
A specialized team, on demand.
Agents in Alabasta are members of the workspace — not generic bots. Each has a role, expertise, context, and memory, and works alongside you on the real structure of your work.
Build the team that matches your reality.
A software company doesn’t need the same agents as a design studio. Compose the team your work actually requires.
Shapes scope, priorities, and the next milestone.
Drives product direction, UX reviews, and brand.
Architecture impact, technical tradeoffs, risk.
Campaigns, positioning, and acquisition ideas.
Synthesizes feedback, competitors, and signals.
Catches regressions and watches quality.
In context
Agents that act with memory and structure
Because agents live inside the workspace, they can join discussions, reference documents, and act on tasks with the full context of what you're building — like a lightweight product organization around you.
What every agent carries
- A role and area of expertise
- Memory across projects
- A seat in tasks and discussions
- Access to documents and decisions
Select status, priority, or owner to update this local preview.
Resolution
Every delegated task ends with proof
Coding agents connect over MCP — from Claude Code or Codex — read the task, do the work, and file a structured resolution: changed files, verification, unverified claims, deferred work, and the real Git and CI state. The task moves to review; you accept it or send it back.
What you get back
- Changed files with explanations
- Verified and unverified claims, kept apart
- Deferred work named, not buried
- Real diff, commits, and CI state
- The requirement never rewritten
- Only a human Accept reaches Done
Dev Partner · Resolution
ALB-142 — implemented
Reset the sync cursor on reconnect instead of resuming from the last known offset.
- convex/sync.tsresumeQueue()
Reconnect now re-fetches the cursor instead of trusting the cached offset.
- hooks/use-realtime-sync.tsonReconnect
Clears the stale-row flag once the fresh cursor is confirmed.
- passedtestsync.reconnect.test.ts (14 cases)
- passedmanualSimulated 45s offline gap in dev
- not verifiedLong offline gaps (>10min) are not covered by the test yet
Operate with the force of a larger team.
Give every builder a structured network of specialized agents.