leanctx 0.0.0

Drop-in prompt compression for production LLM applications. Cut Anthropic / OpenAI / Gemini bills by 40-60%.

leanctx

Drop-in prompt compression for production LLM applications. Cut Anthropic / OpenAI / Gemini bills by 40–60% without changing your code.

# before
from anthropic import Anthropic

# after
from leanctx import Anthropic  # same interface, compressed requests

Open-source models. No API keys to anyone but your existing provider. Your prompts and user data never leave your infrastructure by default.

---

Status: v0.0.0 — name reservation. v0.1 (working release) coming in ~4 weeks. Watch the repo to be notified.

Who this is for

You're building a production LLM app and your token bill is a line item:

• RAG apps with large retrieved documents
• Long-running conversational agents
• LangChain / LangGraph / CrewAI workflows with growing tool chains
• Document-processing pipelines
• Anything where input tokens accumulate and you pay for every one

If your code calls anthropic.messages.create() or openai.chat.completions.create() in production, this is for you.

How it works (coming in v0.1)

Three compression modes, one config switch:

• local — runs Microsoft's open-source LLMLingua-2 locally. Free marginal cost.
• self_llm — lets your own configured LLM do the compression. Highest quality.
• hybrid (default) — routes by content type: code stays verbatim, prose goes through LLMLingua-2, long important spans fall back to self_llm.

Content-aware routing means code blocks, diffs, stack traces, and tool schemas are preserved verbatim — no corrupted syntax.

Roadmap

• v0.1 — Python SDK, local mode (LLMLingua-2), Anthropic + OpenAI drop-in clients
• v0.2 — self_llm mode, Gemini client, LangChain / LlamaIndex integrations
• v0.3 — TypeScript SDK, Docker image, OTel observability
• v0.4 — Helm chart, Kubernetes sidecar deployment

Install (placeholder)

pip install leanctx  # not yet functional — reservation only

Credits

Architecturally inspired by OpenCode DCP (AGPL-3.0, not copied). leanctx is a clean-room implementation under MIT.

Ships Microsoft's LLMLingua-2 (MIT).

License

MIT. See LICENSE.