Skip to content

Single-Agent Workspace

🇹🇭 ภาษาไทย

Not every project needs a whole fleet. BWOC has a single-agent mode: a workspace built around one agent instead of the multi-agent fleet default. This chapter covers when to use it, how to create it, and how it relates to the fleet.

Terminology: BWOC agents always live inside a workspace. There is no machine-wide "global agent" concept — a one-agent setup is a single-agent workspace. If you want one agent reachable from other workspaces/machines, that is the cross-workspace story (peers / A2A), not this one.

When to use single-agent

Use single-agent when… Use the fleet (default) when…
One focused assistant for one project/repo Many agents with distinct roles
Simple setup, no teams/coordination You need teams, shared task lists, peer review
Learning BWOC / a quick experiment A security team, pipelines, division of labour
CI / read-only inspection of a workspace Ongoing multi-agent operations

Create it

bwoc init ./my-ws --single-agent     # one-agent workspace (profile: single-agent)
cd my-ws

What --single-agent changes vs the fleet default:

  • The workspace profile is recorded as single-agent (the default is fleet).
  • The agents/ directory is scaffolded with single-agent-oriented guidance instead of the fleet README.
  • Everything else is a normal workspace: .bwoc/workspace.toml, the agents.toml registry, memory, etc.

For CI / inspection-only workspaces that never spawn a daemon, add --no-runtime (omits the daemon-ephemeral .gitignore patterns; bwoc check still passes). It composes with --single-agent.

Add and drive your one agent

bwoc new sage --role "project assistant" --target agents/agent-sage
bwoc check agents/agent-sage

Then use the full single-agent toolset — no team/peer needed:

bwoc chat agent-sage                     # interactive
bwoc run  agent-sage --task "สรุปงานวันนี้"   # headless
bwoc spawn agent-sage --backend ollama   # pick any backend
bwoc memory put note "..."; bwoc memory search "..."   # workspace memory
bwoc status ; bwoc log agent-sage -f      # observe

Authoring its identity (persona / mindsets / skills) and memory works exactly as in the Agents and Persona·Mindsets·Skills chapters — a single-agent workspace just has one of them.

Relationship to the fleet

A single-agent workspace is the same on-disk structure as a fleet workspace — the flag mainly sets the profile and the scaffolded guidance. That means:

  • Grow later: add more agents with bwoc new whenever the work outgrows one agent; teams (bwoc team) and shared tasks (bwoc task) are available the moment you have more than one.
  • Stay lean: if one agent is enough, keep it — Mattaññutā (right amount): don't add a fleet you don't need.
  • Share across machines: to expose your one agent to other workspaces, use bwoc peer and the A2A protocol — that's federation, not a "global" install.

See also