Every session starts from zero
Yesterday’s findings are gone. The agent re-reads the same sources, re-derives the same conclusions, and bills you twice.
Use case · Research agents
Agents that stoprepeating themselves.
LeanCTX gives research agents persistent, citable memory: findings land in a knowledge store with provenance, the EvidenceLedger keeps verbatim quotes with sources, and cached re-reads cost ~13 tokens. Your agent collects once, remembers across sessions, and answers with citations instead of re-crawling.
Same engine. Different workflow. Always the right context.
What it costs you today.
Answers without receipts
A summary without sources is a liability. Research output needs verbatim quotes with provenance.
Sources drown the window
Twenty open PDFs do not fit in a context window. Agents need ranked recall of what matters, not everything at once.
The capabilities that do the work.
Everything below ships in the open-source binary today. No roadmap items, no waitlists.
Your tools→LeanCTX→Model
EvidenceLedger
verbatim quotes with source attribution, ready to cite
Knowledge store
findings, decisions and blockers persist across sessions
facts / quotes / transcript modes
sources collapse into attributable units
Semantic + BM25 recall
meaning-based retrieval over everything collected
~13-token re-reads
revisiting a known source is nearly free
From zero to first gain.
terminal
# collect a source as quotable evidence
$ ctx_url_read("https://arxiv.org/abs/2310.08560", mode="quotes")# remember a finding
$ lean-ctx knowledge remember "accuracy falls 98→64% with window noise" --category discovery --key context-rot# recall across sessions
$ lean-ctx knowledge recall "context rot"# semantic search over the corpus
$ ctx_semantic_search("why does accuracy fall with window size")One guide. Two journeys. Full reference.
01
guide · Memory concepts
Sessions, knowledge categories and recall.
Learn more02
journey · Web research journey
From sources to a cited answer.
Learn more03
journey · Memory & knowledge journey
What persists, where, and how to query it.
Learn more04
reference · Knowledge tools
ctx_knowledge, ctx_semantic_search and friends.
Learn moreQuestions teams ask before adopting.
The EvidenceLedger stores verbatim quotes with their source URL or file. Answers can reference evidence IDs, so every claim traces back to an original you can open.
Yes. Findings live in a local knowledge store with categories (discovery, decision, blocker…) and are restored into every new session automatically.
Collected sources are deduplicated, compressed and indexed with BM25 plus a knowledge graph. Recall returns ranked, budgeted context, never the whole corpus at once.
Take back control of your context.
Free for local use, forever. CI enforces it. One binary, ten minutes to the first measured gain.