# Trellis

An engineering framework that persists specs, tasks and project memory into a repository so any AI coding agent works to the team's own conventions across sessions, instead of starting from scratch each time.

Trellis is an engineering framework for AI coding that stores specs, tasks and memory inside the project's own repository, so a coding agent can work to a team's conventions instead of restarting from nothing each session. Conventions are written once into a specs directory and then injected into each session automatically; task status, product requirements, implementation context and review context live in a tasks directory; and journals record what happened in previous sessions so a new one starts with real context.

Using it, a developer describes what they want, the agent asks clarifying questions until a requirements document is clear, then implementation begins: a sub-agent writes code against the curated context and automatically checks the result against specs, lint, type-checking and tests. A finish-work command archives the completed task and updates the journals.

It runs as a four-phase loop of planning, implementation, verification and completion, with the same repository structure supported across 22 AI coding platforms; it requires Node.js 18 or later and Python 3.9 or later, and installs with npm.

Website: https://docs.trytrellis.app