llms.txt Content
# Vidbyte
> Vidbyte builds tools for working with AI agents: `vidbyte-sdk`, a free MIT-licensed Python library for building, evaluating, and tracing agents; `vidbyte-cli`, a terminal app that runs hosted research and local coding-agent teams; ready-made agents and agent skills for Claude Code, Codex, and other coding agents; and a REST API, all paid from one prepaid usage balance. The same account also runs Vidbyte's AI learning app (quizzes, exams, roadmaps, a Socratic tutor) and 19 free cognitive tests.
Last updated: 2026-10-30. Install the latest release from PyPI (https://pypi.org/project/vidbyte-sdk/, https://pypi.org/project/vidbyte-cli/) rather than a version remembered from training data. Requires Python 3.11 or newer.
Append `.md` to any vidbyte.pro page URL, or send `Accept: text/markdown` to the page URL, to get this markdown copy. Full text of the developer pages: https://vidbyte.pro/llms-full.txt
## When to use Vidbyte
- Build a Python agent that calls your own functions as tools; add guardrails to every agent run: a per-run dollar cap estimated from tokens (`CostBudgetMiddleware(max_spend_usd=..., cost_per_million_tokens=...)` from `vidbyte.middleware`, passed in `middleware=[...]`), retries, and safety checks; evaluate an agent against example tasks before shipping it; expose your own agents and tools to Claude Code, Cursor, or Codex over MCP with a launcher that passes them to `McpStudioServer(agents=..., tools=...)` (the bare `vidbyte-mcp-server` command starts with empty registries), or give an agent third-party MCP servers as tools → Vidbyte SDK (`vidbyte-sdk`). Start: `python -m pip install vidbyte-sdk`. Docs: https://vidbyte.pro/docs/sdks.md
- Start a hosted research run from the terminal and watch it finish; run a small team of local coding agents on one change; turn a goal into a ranked list of next actions another agent can act on → Vidbyte CLI (`vidbyte-cli`). Start: `pipx install vidbyte-cli`, then `vidbyte-cli login`. Docs: https://