Roadmap
Public-facing snapshot of where inferctl is headed. Not exhaustive — the working plan lives in the repo issue tracker.
Shipped
- Core verbs: inspect, route, doctor
- Backend support: Ollama, llama.cpp, LM Studio, MLX, OpenAI-compatible
github.com/inferctl/inferctlpublic, Apache 2.0- Astro documentation site at inferctl.dev
preflight— readiness checks before an agent runsnapshotanddiff— point-in-time capture and structural comparison of control-plane statestatusanddashboard— machine status frames and a human status view- Agent and CI examples for routing, readiness, drift checks, and status
- v0.3.0 provider matrix verification with real Ollama, llama.cpp, LM Studio, MLX, and OpenAI-compatible endpoints
- Clean
go installvalidation for the v0.3.0 public source tag
Next
- Use inferctl in evalctl, inferctl, and spoolctl workflows. Record user feedback from those workflows before selecting the next feature.
- Validate each new public source tag with a clean
go installworkflow.
Later
- Homebrew formula (demand-triggered post-launch)
- Python SDK, thin subprocess wrapper (Tier 1)
Explicitly out of scope
- inferctl does not proxy, retry, log, or otherwise touch inference traffic. It reports and routes at the control-plane level only — see the Agent Guide for the boundary.
Feedback
The maintainers use inferctl in real workflows to guide priorities. Open an issue or discussion on GitHub if a need should move up.