Lintel
AVAILABLEThe company-neutral session harness. Public beta; this reference follows the 0.12.0 unreleased source.
Open the Lintel repository ↗WORKFLOWS · MEMORY · VERIFICATION
“We agreed on this three hours ago.”
Your AI doesn’t have to start over every time.
Lintel gives your coding agent a reusable way to scope, build, review and remember the work. In the tools you already use.

✓ Intent recovered
✓ Decisions available
→ Next useful action
Start light. See the system. Inspect the evidence. Take the optional deep dive when you want the file paths.
Meet the system, follow one piece of work, preserve its context and test whether the method earns its place.
Lintel, CAIP Pack, Core and Benchmark. Four different jobs, one connected approach.
Open the skills, review records, storage and client adapters. Follow the evidence through a fresh session.
The files are the product. Follow the connections, see where knowledge lives and check what is actually wired up.
Browse the full technical reference →Current source: 49f2d152 · 0.12.0 unreleased. Four client families; 15 starter wrappers. One Review Method, with optional MARS. Discovery files are not proof of live execution.
Service House: a real dinner-shift workflow.
32 frozen requirements. 100 possible points.
Protocol v2 and exact inputs are available.
No completed controlled three-arm comparison is published.
Explore the full harness today. See the planned customer shell and evaluation package. Measure whether the method earns its place.
Download the presentation kit ↓Offline slides · kitchen brief · speaker notes
Website source & update instructions ↗
The company-neutral session harness. Public beta; this reference follows the 0.12.0 unreleased source.
Open the Lintel repository ↗MS Employee in CAIP. The role and organization context, kept separate from the universal core.
A stripped-down shell to build a specific customer harness. Explicit, on-demand workflows inside the customer’s existing coding app — including desktop GitHub Copilot.
For human-driven cases that don’t need a hosted agent or a dedicated app. Hosted execution and platform controls still have their own role.An evaluation package to compare accepted outcomes, interventions, recovery and cost. The kitchen protocol is available above; comparative results and distribution are pending.