Agent Skill
Tuned Tensor exposes an agent-readable skill.md file so coding agents can run local or managed fine-tuning and inspect paired baseline-vs-tuned results.
Fetch the canonical skill file from /skill.md.
Quickstart for agents
The skill tells agents to use tt-local for local-first training and evaluation on a compatible NVIDIA GPU on Linux, and to use tt when managed compute is preferred.
curl https://tunedtensor.com/skill.md
npm install -g @tuned-tensor/local
tt-local init --name "Customer Support Bot" --model Qwen/Qwen3.5-2B --profile spark
# Edit tunedtensor.json, then:
tt-local doctor tunedtensor.json --config local-runner.json
tt-local run tunedtensor.json --config local-runner.json
tt-local runs report <run-id> --config local-runner.jsonFor the account-backed path, see the Managed Quickstart.
Why this exists
Tuned Tensor is designed around a command-line workflow. An agent can manage the same source-controlled tunedtensor.json spec a developer would edit by hand, choose local or managed compute, and inspect the paired run report before iterating.
See also
- Local Training - local-first setup and configuration
- Managed CLI - command reference and configuration
- Managed Quickstart - account-backed spec-to-run workflow
- Authentication - API keys and session auth
- Behaviour Specs - local spec schema