CLIO Coder
GitHub Install CLIO Coder
Menu

CLIO Coder / Tutorials

Get a feel for CLIO Coder.

Short, practical guides to working with CLIO Coder. Start with a real repository, choose your workspace, and learn by doing.

Path 1 / Coding agents

You already use a coding agent

Run a first session in your own repository, see what CLIO Coder may do without asking and how to undo it, check a change against the tests you name, hand bounded work to the agents you already use, then leave a run unattended within limits.

  1. 01Your first session with CLIO Coder5 min
  2. 02Let CLIO Coder work without watching every call7 min
  3. 03Check a change against your repository5 min
  4. 04Coordinate the coding agents you already use6 min
  5. 05Run it unattended within limits6 min

Path 2 / Science and HPC

You write research code and run it on a cluster

Start in your own repository, follow a numerical fix from a failing test to recorded verification, let the agent submit and follow a Slurm job you asked for, keep a speed-up only when a measurement proves it and the answers still match, and coordinate workers on other machines over SSH.

  1. 01Your first session with CLIO Coder5 min
  2. 02A numerical change, checked with CLIO Coder5 min
  3. 03Submit a Slurm job through CLIO Coder5 min
  4. 04Run a research campaign and a blind arena7 min
  5. 05Work from a laptop with remote workers11 min

Path 3 / Local models

You run your own models

Connect a model server with no account or key, choose models for conversation and for delegated work, read what each session cost and why a run stopped, carry a long task through a handoff without losing its decisions, then compare answers blind before you trust a ranking of your models.

  1. 01Run CLIO Coder on a local model4 min
  2. 02Choose a model for your project6 min
  3. 03See what a session cost and what a run did4 min
  4. 04Keep a long coding task on track4 min
  5. 05Run a research campaign and a blind arena7 min

All tutorials

Terminal recording of a first launch of CLIO Coder 0.8.0 with no model configured: clio-coder starts setup, which finds a LiteLLM model server at http://127.0.0.1:4000 serving qwen3.8-27b. Choosing it sends one test message, answered in 0.4 s, and saves the settings. Chat opens with the model ready, the first prompt asks how the repository builds and runs its tests, and the recording ends while the agent plans its first reads of the repository.

Getting started / 5 min

Your first session with CLIO Coder

Connect a model, open a repository, and ask CLIO Coder to explain how it builds and runs its tests.

Read the tutorial CLIO Coder desktop after a completed calibration task: seven passing tests and the reference temperatures in the conversation, with task status, workspace, model, context and usage in the Session column.

Product tour / 4 min

Desktop or terminal: find your workspace

See both interfaces, learn how to launch them, and choose where to start your next coding session.

Read the tutorial Terminal recording: clio-coder doctor's shell sandbox row lists the writable roots and masked secret files. On a scratch branch the agent deletes dist/, rebuilds it with node build.mjs, and runs the declared test check with 7 passed. /rewind lists the snapshot taken before the delete; its preview says the workspace already matches and gitignored files are never touched, and Esc leaves without restoring.

Safety and autonomy / 7 min

Let CLIO Coder work without watching every call

Check the sandbox, undo a change with /rewind, tell the safety judge about your machine, and know when yolo is the right call.

Read the tutorial

Workers and fleets / 5 min

Run a playbook from plan to commit

Run the builtin sdlc playbook on one task, watch it work, and read the three commits it makes on your branch.

Read the tutorial Terminal recording of clio-coder run --json writing its events to run.jsonl and exiting 0, jq printing the final answer that 6 of 7 tests pass with the absolute-zero test failing, and the last event, run_end, with outcome succeeded and safety decisions of 1 allowed, 0 blocked and 0 permission requests.

Safety and autonomy / 6 min

Run it unattended within limits

Start a headless run with a cost ceiling and a timeout, ask read-only questions, schedule turns, and know what stops each one.

Read the tutorial

Accountability / 4 min

See what a session cost and what a run did

Read session cost and stop reasons from local journals, verify a sealed record, and prepare an archive that uploads nothing.

Read the tutorial Terminal View dock with the session's records on the left and the checked test row selected; the preview shows check test, its working directory examples/temperature-calibration, exit 0, the test command, and the passing test lines.

Scientific software / 5 min

A numerical change, checked with CLIO Coder

Follow a temperature-conversion example through a floating-point test failure, a correction, and recorded verification.

Read the tutorial Guided setup after choosing a model server and entering its address: the server is reachable with one model, qwen3.8-27b is taken as the only one listed, the test message is answered, and the connection is saved as the chat and worker default, with a note that the server reported no context window.

Model choice / 6 min

Choose a model for your project

Connect a local server, a cloud API, or a supported subscription route, then choose models separately for conversation and delegated work.

Read the tutorial Terminal Local coding agents view: Claude Code, Codex, OpenCode and GitHub Copilot proposed for connection and not configured, their inventories with nothing to adopt, Antigravity CLI not installed, and the configuration entry that connecting Claude Code would append.

Agent coordination / 6 min

Coordinate the coding agents you already use

Use supported coding peers for bounded assignments, distinguish managed runs from interactive handoffs, and review the results before combining them.

Read the tutorial Terminal Context Window dock after one short turn: a usage bar with the tokens in use out of the 200,000-token window, split into system prompt, tool definitions, agent fleet, project context, messages, tool results and assistant output, then free space, the autocompact reserve, and the next request's input, output reserve and headroom.

Context and memory / 4 min

Keep a long coding task on track

Inspect context, preserve decisions through a handoff, and resume the correct branch without turning every past message into a current instruction.

Read the tutorial Terminal turn ending with CLIO Coder's report: the cause was kelvin <= 0, which rejected 0 K; changing it to kelvin < 0 at line 21 of temperature.mjs keeps negative Kelvin rejected, and all seven tests now pass.

Verification / 5 min

Check a change against your repository

Declare which checks matter, run them through verification, and inspect whether passing results still match the files you are reviewing.

Read the tutorial Terminal Library Installed view at project scope with the tdd skill available and trusted: how to use it with /skill tdd, its kind, source and origin, and the keys to disable, update or remove it.

Skills and plugins / 5 min

Add skills and plugins deliberately

Preview a project skill or plugin, inspect its package pin, and load only what your task needs; installing is not a permission grant.

Read the tutorial Terminal Settings open on Workers with Add SSH node selected under Machines: automatic routing, the ready local node, Tailscale discovery, workers at once, and fleet retry and tool limits.

Remote workers / 11 min

Work from a laptop with remote workers

Use a local CLIO Coder session to coordinate SSH workers when your machines share the required project path, runtime version, and approved access.

Read the tutorial

Extend CLIO Coder / 5 min

Write an extension with declared capabilities

Scaffold, test, and install a small extension, and see exactly what approving its capabilities covers.

Read the tutorial

Models / 3 min

Diffusion models in the terminal

Connect Inception's Mercury models for chat and workers, and watch an answer refine as a whole instead of arriving token by token.

Available since 0.5.3, terminal display only

Read the tutorial

Terminal workspace / 5 min

Panes and docks: files, workers, and music

Show and hide a file browser, a live worker dashboard, and a focus radio beside the terminal conversation with one key each.

Experimental, off by default, needs Herdr

Read the tutorial

Safety / 3 min

Keep named files with the models you chose

Declare which paths and tool results may reach which model targets, approve the policy, and see a request to any other destination refused.

Advanced, opt-in

Read the tutorial

Context / 6 min

Search by meaning and record worker runs

Index code, memory, and experiment folders with a local embedding model, and attach a redacted terminal recording to a worker's evidence.

Beta, off by default

Read the tutorial

Model choice / 4 min

Run CLIO Coder on a local model

Connect a local server with no account or key, check what CLIO Coder learned about it, and see what else to check if code must stay on your machine.

Read the tutorial

Scientific software / 5 min

Submit a Slurm job through CLIO Coder

Declare the clio-kit Slurm server, install the skill, and let the agent submit and follow a job, with a confirmation before it reaches the scheduler.

Read the tutorial

Scientific software / 5 min

Pre-register a benchmark before you measure

Use the experiment-protocol skill to lock thresholds, tolerances, and verdict conditions into a contract you commit before any result exists.

Read the tutorial Terminal: store put saves run-042.csv under the results tag with a note, then store stat shows its ref, path in the store, size in bytes, modification time, drift none, its sha256, no producing run on record, and the local backend.

Scientific software / 5 min

Keep results outside the tree with the project store

Store a result with a note, check it for drift, read it from a script, and let a worker write into one granted tag.

Read the tutorial

Models / 5 min

Set up a lab from one setup file

Write one YAML file that connects every lab member to your model gateway, check what it may carry, and test the route.

Read the tutorial

Terminal workspace / 4 min

Share a folder with a teammate or a second terminal

Run two sessions in one project, keep each person's state private, and commit project settings that can only make CLIO Coder stricter.

Read the tutorial

Scientific software / 7 min

Run a research campaign and a blind arena

Keep a speed-up only when a measurement proves it and the answers still match, then compare model answers without knowing which model wrote them.

Read the tutorial

Accountability / 5 min

Report a problem without leaking your machine

Draft a bug report, read the exact text with your paths, names and keys masked, choose how it leaves, and share a diagnostic bundle only if you want to.

Read the tutorial

Agent coordination / 5 min

Ask several agents at once

Poll two to five agents on one question and read the vote, or give a council roster one task and let a judge write a single answer.

Read the tutorial