Frequently asked questions about x-dog.
What is x-dog?
x-dog is a Python monorepo of composable packages for building AI agents and managing LLM deployments: a unified provider API (ai), an agent runtime (agent), a terminal UI library (tui), an interactive coding CLI (coding), an orchestration runtime with long-term memory (claw), and a typed local-first workflow format/compiler for humans and Coding Agents (flow).
Which package do I need?
Start with ai if you just want one interface over many model providers. Add agent to get a tool-calling loop. Reach for flow when a developer or Coding Agent needs to crystallize a repeatable process as validated, Git-friendly JSON, then run, compile, or schedule it locally. tui is the rendering layer; coding and claw are ready-to-use applications built on the rest.
Which model providers are supported?
Through the ai package: Anthropic, OpenAI, Google, Amazon Bedrock, Mistral, and GitHub Copilot. Application code selects a model by id; switching providers does not require changing the calling code.
How does flow's code generation work?
A workflow is authored as JSON. You can execute it directly with the runtime, or run xdog-flow generate to compile it into a single self-contained Python module. The generated code keeps node outputs in the same nested port structure the interpreter uses, so both forms behave identically, and it passes the same ruff and mypy --strict checks as hand-written code.
How do I add a custom tool to a flow agent?
Declare it in the workflow's top-level tools manifest as a name mapped to a module:func reference, then list that name in an agent node's tools. The reference is imported at run and generate time (the workflow's own directory is on the path, so you can bundle tool files next to the JSON) and registered under the manifest name. Unknown tool names fail validation early.
Can I see and edit a workflow visually?
Yes. xdog-flow build opens an interactive terminal builder with graph, node/edge, Functions, and Tools views. A local Web UI is planned as a Workflow JSON IDE; both UI surfaces and Coding Agents edit the same file rather than a database-only model.
Is flow production-ready?
Flow ships typed ports, explicit mappings, frontier-based parallel execution, retry policies, coherent batch checkpoints, human signals, failure isolation, dynamic fan-out, subflows, standalone codegen, and local scheduling. It is production-appropriate for single-machine developer automation, but deliberately not a distributed durable-execution platform or multi-tenant hosted service.
How do I install and run it?
Clone the repository and use uv: 'uv sync' installs all workspace packages, then the console scripts are available via 'uv run' — xdog-ai, xdog-agent, xdog-coding, xdog-flow, and xdog. Python 3.11+ is required.
What is the license?
x-dog is MIT-licensed.