Step 01
Ingest and map
Tree-sitter parsing builds deterministic symbols, call-sites, and query indices across the repo.
General Landing
Mesh is one product family with three equal surfaces powered by one structural context engine.
Step 01
Tree-sitter parsing builds deterministic symbols, call-sites, and query indices across the repo.
Step 02
Files are converted into structural capsules so models can see more code at the same token budget.
Step 03
When exact bytes are required, Mesh serves precise source ranges instead of re-sending whole files.
Step 04
Context is delivered through IDE, MCP, or API with shared semantics and predictable budgets.
01
Browser runtime with editor, terminal, AI chat, review context, and voice-assisted coding loop.
02
Standalone MCP surface exposing the same context engine for external agents and local workflows.
03
Programmable integration surface for automations, product features, and backend orchestration.
Docs / Benchmarks
Compression Ratio
NIAH Recall
Hallucination Probe FPs
MCP Compatibility