Raw JSON goes straight into the model
Every tool response, API payload and document lands unprocessed in the prompt. You pay for noise on every single call.
Use case · Custom bots
Give your botcontext engineering.
LeanCTX embeds in any custom bot or agent via a versioned /v1 HTTP API (OpenAPI + /v1/capabilities) with GA Python and TypeScript SDKs plus a Rust SDK. Your bot gets compressed reads, persistent memory, a knowledge graph, the ctx_tools gateway and governance, without building any of it. One local Rust binary, lean-ctx serve, zero cloud dependency.
Same engine. Different workflow. Always the right context.
What it costs you today.
Home-grown memory is fragile
A vector store bolted to a cron job is not a memory system. Findings vanish, sessions reset, and nobody trusts the recall.
Every bot rebuilds the same plumbing
Caching, deduplication, retrieval, budgets: each new bot reinvents the context stack instead of shipping its actual job.
The capabilities that do the work.
Everything below ships in the open-source binary today. No roadmap items, no waitlists.
Your tools→LeanCTX→Model
Versioned /v1 API
OpenAPI spec + /v1/capabilities discovery; read, search, memory and shell endpoints
GA SDKs + conformance
Python & TypeScript SDKs GA with a 14-check conformance kit, plus a Rust SDK
ctx_tools gateway
bridge external MCP servers and plugins through one governed entry point
WASM extensions
a sandboxed WASM runtime and plugins extend the layer without forking it
Context personas
built-in personas (coding, research, support…) plus custom; one switch reshapes tool set, read mode, compression and intents
Knowledge store
facts, decisions and blockers persist across sessions and restarts
From zero to first gain.
terminal
# start the context API
$ lean-ctx serve# discover the surface (OpenAPI + capabilities)
$ curl localhost:8080/v1/capabilities# install the SDK
$ pip install lean-ctx-client# connect from your bot
$ client = LeanCtxClient("http://127.0.0.1:8080")# compressed read via the /v1 API
$ client.call_tool_text("ctx_read", {"path": "data/report.pdf"})One guide. Two journeys. Full reference.
01
guide · API reference
Every /v1 endpoint with request and response shapes.
Learn more02
journey · Open door: SDKs & API
Embed LeanCTX in your own agent, end to end.
Learn more03
journey · Context personas
Tune the layer for your bot’s domain.
Learn more04
reference · MCP tool catalog
The full 81-tool catalog your bot can call.
Learn moreQuestions teams ask before adopting.
No. lean-ctx serve runs a versioned /v1 HTTP API on your machine. Any process (a Slack bot, a support agent, a pipeline) can use compressed reads, memory and search via the Python, TypeScript or Rust SDK.
Yes. The Python and TypeScript SDKs are GA against the frozen /v1 contract, and a 14-check conformance kit keeps every client honest. A Rust SDK and an OpenAPI spec (/v1/openapi.json) round out the surface, and extensions run as sandboxed WASM or plugins so you extend the layer without forking it.
On your machine. LeanCTX is local-first with zero telemetry: cache, knowledge store and ledger are local files. Nothing leaves unless you build something that sends it.
Nothing. Local use is free forever, enforced by a public CI gate. Paid plans only add hosted sync and team planes on top.
Take back control of your context.
Free for local use, forever. CI enforces it. One binary, ten minutes to the first measured gain.