x-dog

Tools for building AI agents and managing LLM deployments.

GitHub →

x-dog / Packages / claw / Reference

The top-level API, the workspace files that shape the prompt, the xdog-claw CLI, and the input queue modes.

Top-level API

Exported from claw/__init__.py.

Name Kind Purpose
Orchestrator class Single dispatch path over all groups.
GroupRuntime class One group's workspace, memory, goals, tools.
AgentSession / TurnResult class A running turn and its result.
MessageQueue / TaskScheduler class Input queueing and scheduled jobs.
TranscriptStore / SessionManager class Persistence of conversations.
build_system_prompt / init_workspace function Assemble the prompt / seed the workspace files.
ClawConfig / GroupDef type Global and per-group configuration.

Workspace files

The files under a group's workspace that shape the system prompt. Editing them (or running onboard) changes the agent.

File Role
IDENTITY.md Who the agent is — the Name line is the agent's name.
AGENTS.md Operating instructions / behavioural rules.
SOUL.md Persona and voice.
USER.md What the agent knows about the user.
MEMORY.md Long-term memory snapshot.
BOOTSTRAP.md One-time first-run setup, consumed then deleted.

CLI — xdog-claw

Subcommands of the claw console script (cli/cli.py).

Command Does
onboard Setup wizard: provider login, model, agent name, workspace.
gateway start [--config] [--foreground] Start the gateway daemon.
gateway stop · gateway status Stop / inspect the gateway.
tui [--group] Attach a terminal client to a group.
channel login --weixin [--base-url] Authenticate the WeChat channel.

Queue modes

How a burst of input to one group is folded together.

Mode Behaviour
collect Batch pending inputs into the next turn.
steer Interrupt the current turn with new input.
steer-backlog Steer, keeping a bounded backlog of the rest.