Tuned Tensor

Agent

The tt agent runs on your machine and helps you prepare workflows and inspect results. Coding agents can also use TT through the instructions in skill.md.

Interactive shell

TT runs tools, handles approvals, and stores conversations locally. To use the managed model, save an access token from your dashboard settings once:

tt auth login
tt

If you have not selected a provider, this selects the managed model automatically. Tuned Tensor chooses the model and sends requests through OpenRouter. The same token gives you access to cloud operations, with no separate OpenRouter key to configure.

To use your own OpenRouter account and model, open TT and select it:

tt
› /login openrouter
› /model openrouter/<model-id>

Requests go directly to OpenRouter with your key and chosen model. Use /model tunedtensor/managed to switch back, or /login tunedtensor to save a Tuned Tensor token in the shell. Direct local commands require no token. Either model option works with local or cloud workflows. Cloud operations require a TT token.

Optional provider keys are stored in ~/.tuned-tensor/agent/auth.json (or $TUNED_TENSOR_HOME/agent/auth.json). When TUNED_TENSOR_HOME is unset, existing XDG agent state remains in use until the new agent directory exists. Prompts and tool results go to Tuned Tensor and OpenRouter for managed inference, or directly to your chosen provider when using your own key.

Ordinary sentences go to the agent. Known commands such as runs list and doctor execute directly. Prefix a command with : to make that intent explicit.

› What happened in my latest training run?
› runs list

The model has no shell or general filesystem tool. It can prepare a new folder beneath the current working directory with a validated tunedtensor.json. Spec writes wait for /approve, require at least two examples, and fail if the path already exists. /reject discards the proposal. The agent can also prepare a pipeline dry run for approval. Start or cancel actual training with a direct CLI command. With cloud access configured, the agent can propose creating or updating a cloud spec for approval in TT.

Cloud spec approval requires the same TT token and API origin used to prepare the proposal. After changing either, restore the original token and API origin or prepare a new proposal.

Use /model to inspect or change the TT agent model. Set the training base model separately in your spec. List providers with /model, list one provider with /model <provider>, search with /model <query>, or switch with /model <provider>/<model>. Scripts can use tt agent configure after the relevant credential is saved. Other controls: /help, /status, /context, /login, /new, /threads, /resume, /cd, /clear, /exit.

Skill for coding agents

Read the instructions for coding agents at /skill.md. It tells coding agents to install tt once and run the same local and cloud workflows.

curl https://tunedtensor.com/skill.md
curl -fsSL https://tunedtensor.com/install.sh | sh
tt init --name "Customer Support Bot" --model Qwen/Qwen3.5-2B --profile spark
# Edit tunedtensor.json, then:
tt doctor tunedtensor.json --config local-runner.json
tt pipeline run --spec tunedtensor.json --config local-runner.json
tt runs report <run-id> --config local-runner.json

Working with a spec

Keep tunedtensor.json in version control. Review changes to its examples and instructions, then use the run report to decide what to change next.

See also