Skip to content
BMO

Decision guide

This guide answers “which path?” for evaluators and new operators. It explicitly marks what to ignore on day one.

Start here (ignore the rest until you need it)

Section titled “Start here (ignore the rest until you need it)”
Your jobStart withIgnore for now
Fix one bug or small featureQuickstart → Common WorkflowsMesh, stigmergy, teams, choreography
Daily interactive codingTUI + Session modesArena, shadow evals, fleet metabolism
CI or cron automationAutomation + SandboxMulti-agent, workstreams
Editor integrationMCP Server or ZedAgent mesh, A2A
Compare models on one promptArenaQuality orchestration fan-out
Parallel agents in one processMulti-Agent Workflows + workspace claimsStigmergic delivery (research preview)
Remote peer by capabilityAgent Mesh + A2ATeams (local coordination first)
Review before mergeReview before mergeShadow evals

Decision tree

Section titled “Decision tree”
flowchart TD
  start["What are you trying to do?"]
  start --> solo["Solo coding in terminal"]
  start --> headless["Script / CI / cron"]
  start --> editor["IDE or external client"]
  start --> multi["Multiple agents"]
  solo --> tui["bmo TUI or bmo run"]
  headless --> run["bmo run + automation guide"]
  editor --> mcp["bmo service start mcp"]
  multi --> claims["Workspace claims first"]
  claims --> local["Subagents / teams"]
  claims --> remote["Mesh / A2A"]

For coordination surface detail (subagent vs nanite vs team vs mesh), see Agents: kinds and coordination.

Surface picker

Section titled “Surface picker”
SurfaceWhenRisk note
bmo (TUI)Interactive work, approvals, discoveryTool policy and grants determine prompts
bmo runOne-shot, scripts, CIRequests needing interactive approval require an approval listener
bmo service start http|mcp|autopilotLong-lived integrationConfigure service-specific auth and binding; see HTTP server
bmo acpACP consumersSame core, surface-specific policy UX
A2A HTTPPeer agentsRemote trust expansion

Feature depth ladder

Section titled “Feature depth ladder”
  1. Daily-use — TUI, bmo run, file tools, patch proposals, session modes, staged workflow (optional).
  2. Advanced — automation, sandbox, multi-agent, phase model, operating packs.
  3. Advanced (research preview on feature page) — Stigmergic delivery.
  4. Eval-only — shadow evals, session-mode autoselect eval (measurement, not authority).

Full label definitions: Limitations and maturity map.

Section titled “Related”