Trellis is a specification format and tooling layer that pairs each unit of source code with a structured sidecar capturing what the unit is for, what it provides, what it consumes, and what it must not do — a first-class artifact that humans and AI agents can both reason over before a change is made.
Compelling at every level of the stack.
Whether you're a solo developer managing an AI-assisted project or an enterprise architect responsible for system-wide coherence across dozens of teams, Trellis addresses the problems that matter at your level. The stories below are coming.
For solo developers & individual contributors
Keeping a fast-moving codebase coherent when you're the only one in it
Trellis at the unit level: how a sidecar file disciplines an AI agent, prevents duplication and drift, and keeps intent legible as the codebase grows under AI-assisted pressure.
This story is coming.
For engineering teams
Staying aligned when the whole team is shipping with AI assistance
How Trellis changes team-level AI workflows, code review practices, and architectural discipline — addressing the failure modes that compound when many agents operate across a shared codebase.
This story is coming.
For staff engineers & architects
Enforcing the rules that matter most across languages and teams
A uniform sidecar format as a polyglot substrate for architectural enforcement — layer boundaries, bounded contexts, and stability tiers — mechanically, in CI, without relying on reviewer vigilance.
This story is coming.
For engineering leaders & organizations
Scaling AI-assisted development without losing architectural coherence
Trellis as organizational infrastructure: policy packs, ownership-aware change analysis, drift dashboards, and the case for a shared intent substrate across an engineering portfolio.
This story is coming.
Core Thesis
"Tests verify behavior. Code embodies behavior. Trellis preserves intent — as a reviewable, version-controlled artifact that both humans and agents can reason over before a change is made."