Zero deep dive

What Zero is for

Zero is the product surface: authenticated users, organizations (workspaces), threads, streaming answers, and an inspector that shows which agents and tools ran.

Mental model

User message
  → Director (routes / clarifies)
  → Specialist agents (tools + research)
  → Synthesizer (composed answer)
  → UI (markdown + data/chart/reference tabs)

The live Agent run graph grows as SSE trace_update events arrive. That graph is a selling surface — leave it visible while streaming.

Answer modes

Mode ids are fixed: summary, light, extended, project_plan, roadmap.
Crews choose which appear and how they’re labeled via crew.yaml. Semantics (length, structure) are enforced in Zero’s agent prompts.

Rich answers

When the synthesizer cooperates with output_composition:

Block	How it shows up
Summary markdown	Default tab
Tables / JSON	[DATA_START]…[DATA_END] → Data tab
Charts	visualizer tool → Visualizations
Citations	structured extract + References

Workspaces and files

Organizations hold shared files. Agents can search them through knowledge/file tools. That is the primary path for “bring your own data” today — see Bring your data.

Ops essentials

• Load crews with ./scripts/load-crew.sh --restart <crew-dir>
• Prefer a paid LLM tier or set OPENAI_API_KEY — multi-agent + tools burns free-tier RPM quickly
• GET /api/crew exposes the active manifest to the UI

Next

Author crews in CrewDefine, then follow getting started.