Developer Preview

Stop paying AI
to re-read your codebase
every session.

Semantic Cache Engine sits alongside GitHub Copilot and caches your reasoning context locally. Warm sessions answer from cache — zero extra tokens, instant recall.

75–78% session-start token reduction
51× faster on cache hits
0 outbound connections

We eliminate redundant AI compute.

Every token an AI model processes requires compute. Every compute cycle burns energy. Most of that work is redundant — the same context, re-read, re-embedded, re-sent, on every session start.

LeanGrid Labs builds tools that eliminate that redundancy. Not as a side effect of performance optimization, but as the primary goal. Leaner AI usage is faster, cheaper, and less wasteful.

Semantic Cache Engine is the first tool in that line: local-first reasoning context that persists across sessions, travels with your team, and compounds in value the longer you use it.

We're building IDE-agnostic. VS Code is first. Everything else follows.

Local-first
Your reasoning cache lives on your machine. Nothing leaves without you explicitly exporting it.
Provably lean
netstat-verified zero outbound connections during local operations. We show the test, you run it yourself.
Compounds over time
Every session adds to the cache. Every shared bundle starts a teammate warm on day one.

Your AI relearns your codebase from scratch. Every time.

📄
Session starts
AI sends your full codebase context — thousands of tokens — to the model.
🔄
LLM processes it
The model reads context it already understood yesterday. And the day before.
💰
You pay again
Same tokens. Same cost. Every session, every developer, every day.
With Semantic Cache Engine
Session starts
Engine checks the local cache. Context is already there.
📌
Cache hit
Answer returned in ~41ms. No LLM call. Zero tokens used.
🆕
You keep working
75–78% of session-start tokens never leave your machine.

How it works.

01

Install the extension

Drop the VS Code extension into your workspace. No account. No API key. No configuration required to get started.

02

First session indexes your codebase

The engine ingests your project, chunks it semantically, and builds a local reasoning cache in .reason/. One-time cost.

03

Every session after starts warm

The resolver checks the cache before routing any question to the LLM. Cache hits answer instantly — no model call, no tokens, no wait.

04

Share bundles with your team

Export a privacy-scrubbed bundle and share it. Teammates import it and start their first session with the cache already warm.

Recall
Question answered directly from cache. ~41ms. Zero tokens.
Graph-assisted
Graph context reduces the prompt before calling the LLM. ~35% context reduction.
Agent handoff
Novel question — full agent context. Answer is cached for next time.

Claims we stand behind.

75–78%
Session-start token reduction
Measured with real OpenAI API calls (gpt-4o-mini), verified on two separate machines. Engine-on vs engine-off A/B test.
Claim 1b — real API measurement
0
Third-party outbound connections
Verified via netstat on two machines during local operations. The extension includes a built-in no-phone-home check you can run yourself.
netstat-verified
100%
Reproducible demo
Cold-to-warm savings are reproducible in the Docker demo on any machine with a Docker runtime. Run it yourself in under 5 minutes.
Docker demo — try it live
Local
All reasoning stays on your machine
The cache lives in .reason/ in your workspace. Bundles only leave when you explicitly export and share them.
Architecture guarantee
Team savings compound. The 75–78% figure is per-developer, per-session. When teams share bundles, each teammate starts warm on day one. The more overlap in your codebase, the more each person benefits from what everyone else already cached.

Cold session vs. warm session.

Session-start walkthrough
Cold → warm — token reduction shown live
Team bundle import
Export, share, import — teammate starts warm
Docker demo
5-step cold-to-warm proof you can run yourself
docker pull ghcr.io/leangridlabs/semantic-cache-engine-demo:latest
docker run -it --rm ghcr.io/leangridlabs/semantic-cache-engine-demo

Walks through ingest → cold run → export → import → warm run interactively. No account required.

Get notified when the dev tier ships.

The free developer tier is in preparation. Leave your email and we'll let you know the moment it's available — no spam, one email.

No spam. No data sharing. Unsubscribe any time.