Fetching from the wire…
Public story · 2026-08-10 · high
LeanCTX is a Rust proxy, MemPalace hooks in before compaction fires, and claude-mem re-injects sessions, per their repos.
Why now: The three tools surfaced together in coverage dated August 10, 2026.
Three separate projects landed on the same fix independently, per coverage dated August 10, 2026. That matters for anyone running agents through long sessions. Memory handling stops being a prompt technique and becomes a separate process you deploy and trust.
LeanCTX is a Rust binary that proxies the wire and decides what an agent is allowed to read. MemPalace runs local-first with zero API calls and hooks in before compaction fires. claude-mem captures what a session does, compresses it, and re-injects it on the next run, per its GitHub repo.
The shared target is vendor auto-compaction. All three treat it as the lossy step, not the solution. That's a claim about where the failure lives: the vendor's own summarization, not the model's context window size.
A Rust binary proxying the wire is a process boundary. It runs outside the vendor's control, which means it can preserve what the vendor's compaction would drop. That also makes you responsible for running and trusting a separate piece of software to keep your agent's memory intact.
Watch whether one of these gets folded into a vendor's own tool instead of staying independent. That would show whether vendors see local context control as a threat worth absorbing or a feature request.
None of the three report how their latency or context retention compares to vendor compaction, per their repos. That's the gap to check before picking one for production.
Each link below shares sources, entities, or timing with this story.
Headroom compresses tool outputs, logs, RAG chunks, and files before they ever reach the model. It deploys as a library, a proxy server, or an MCP server, and the benchmarks are blunt: 92% token reduction on code search (17,765 down to 1,408) and SRE debugging (65,694 down to...
This is the most honest thing published about agents this year, and it's from a SaaS blog, not a research lab. SaaStr has been running 20+ AI agents in production for a year, going from 8 or 9 human salespeople to 1.2 humans plus 20 agents. Then they published a post-mortem on...
MIT-licensed Rust, created August 27. It stores LLM observations as provenance-tracked facts with stratified rules and incremental derivation, exposed to a harness over MCP, so the model does language extraction while the database keeps logical consistency. Reported: 0.463 ±0....
lean-ctx is a Rust-based system that sits between AI coding tools and LLMs, compressing file reads by 60-99% and shell output by 60-95% using Tree-sitter AST parsing for 18 languages. Cached re-reads cost only 13 tokens. Works as a standard MCP server with 49 tools. Compatible...
claude-mem (89.3K stars) targets seven harnesses. lean-ctx (3.5K stars, Rust, 76 MCP tools) uses 60-90% token reduction as its headline metric. Wienerdog surfaced on HN August 2 offering memory plus self-improving skills for both Claude Code and Codex. All three are harness-ne...
Every story above generates tokens, and tokens are money. rtk is the clearest "do this today" item in the whole dataset. It's a single Rust binary, 59,658 stars, created January 22 and pushed as recently as June 7, that proxies common dev commands and claims 60 to 90% reductio...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.