plur-ai/plur

PLUR is local-first, open shared memory for AI agents: corrections, preferences and conventions persist across sessions, tools and machines as plain-text YAML engrams on your own disk - no cloud, no API calls, no black box. Installable into DeepSeek Harness with dsh plugin add @plur-ai/dsh, it publishes a reproducible LongMemEval-S benchmark harness and reports that a small model with PLUR memory beats a larger one without it at roughly a tenth of the cost.

PLUR is local-first, open shared memory for AI agents. Install it once and corrections, preferences and conventions persist across sessions, tools and machines, stored as plain-text YAML engrams on your own disk that you can read, correct and delete rather than weights you cannot. No cloud, no API calls, no black box. PLUR publishes its retrieval and agent-task benchmarks openly - LongMemEval-S full-corpus R@5 (BM25 92.2%, hybrid 95.6%, plus a local cross-encoder reranker 97.6%), separately-reported end-to-end answer accuracy, and an agent-task result where a small model with PLUR memory outperforms a larger model without it at roughly a tenth of the cost. The reproducible harness lives in plur-ai/plur-bench and is what CI regression-checks.

Memory ★ 300
View on GitHub ↗

Install

dsh plugin add @plur-ai/dsh

Copied verbatim from the README's Install section for DeepSeek Harness. The package is also offered to other MCP-tool hosts; PLUR stores memory as plain YAML on your own disk.

Compatibility

Local-first with no cloud call required on the default path; an optional cloud embedder (openai-3-large) can be enabled. The README lists support across MCP-tool hosts including Claude Code, Codex, Cursor, OpenCode, Hermes and OpenClaw. English README.

Details

Alternatives

863683348/dsh-plugin-focus · akslcw/dsh-negative-ledger · aLIZELI/belief-merge

More plugins in Memory

Browse more in Memory

Guides for Memory plugins