Skip to content

Tutorial

This tutorial walks through every layer of Tether, from running your first local 9B session in the terminal to orchestrating multi-agent workflows with custom Python tools and SQLite FTS5 retrieval.

  1. CLI and Headless Sessions — Run foreground sessions, spawn detached background workers (--headless), inspect live progress (status, wait), send live follow-ups (send), and query results (read, query).
  2. Python SDK and Custom Tools — Configure Harness, decorate Python functions with @tool, pass multiline payloads over stdin, and leverage progressive --help disclosure.
  3. Built-in Tools — Master the opt-in built-in namespaces: done, bash, edit, web, mem, and agent.
  4. Progressive Skills — Bundle domain instructions (SKILL.md) and custom Python tools (tools.py) into reusable skills discovered in ./skills/ or ~/.tether/skills/.
  5. Async Multi-Agent Orchestration — Spawn parallel child sessions (agent spawn), wait for compact 50-token digests (agent wait), and run sub-millisecond BM25 RAG across sub-agent observations (agent query).
  6. Security and Sandbox — Enforce deterministic path confinement, .env/.git protection, secret environment scrubbing, and Linux bwrap kernel isolation.
  7. Gates, Memory, Pipelines & Telemetry — Verification gates (--verify), done --ask, project memory (mem --project), YAML pipelines, tether stats and tether retro.

Throughout the tutorial, the course-builder example (examples/course/, a three-stage pipeline that plans, researches and writes a small course with ollama/qwen3.5:9b) is the running example: each page shows the piece of it that illustrates that feature, plus the limits we hit. Problems are collected in Troubleshooting.