agented или io.github.las7/memharness

A text editor for LLMs, not humans. Optimized for the model's cost model: fewer round trips per task (no Read-before-Edit; conflict responses carry the new content, so you reconcile in one call instead of pre-reading every time), tighter token budget (lean tab output, not JSON envelopes; ~5x denser than the built-in Read/Edit shape), inline LSP diagnostics on every save (gopls, rust-analyzer-with-clippy, tsserver+eslint findings come back as diag lines instead of round-tripping through cargo / go build), reactive drift detection (external editor or formatter writes are folded into the undo tree as new edits, not silently clobbered or lost), and a branching undo tree that remembers the branches the agent abandoned — because that's often where the interesting work was. Three-way merge for concurrent agents, atomic multi-file refactors, cross-session memory via per-file annotations.

Bi-temporal, provenance-carrying long-term memory for AI agents, in one SQLite file.