jinguanghai/deepseek-harness-forge-plugins#forge-memory

BM25 keyword-based memory recall.

BM25 experience recall for DeepSeek Harness, with a folded-memory index attached. memory_recall takes a query and returns ranked snippets from your memory files, so past experience is retrieved by keyword relevance rather than being re-derived. Three companion tools manage folded memory: memory_fold_list lists the folded sections (π/φ index), memory_fold_preview previews a section without pulling it into context, and memory_fold_deep reads a section in full — which lets an agent survey a large memory store cheaply and only expand what it needs.

Memory ★ 0 updated 2026-08-21
View on GitHub ↗

Install

dsh plugin add ./plugins/forge-memory

README INSTALL section form 1 (dsh CLI, local directory from a repo checkout) — or the bundle tarball attached to the GitHub release. The pnpm add dsh-forge-memory form is documented but the npm registry returns 404 for dsh-forge-memory (verified 2026-09-11). The plugin reads memory.json / .forge-temp/memory_current.json / data/memory_current.json; bring your own memory files (empty templates live in data/).

Compatibility

Works with dsh >= 0.1.0-rc.5. Pure Cordis, zero npm dependencies, all file access via ctx.fs. Memory data location is controlled by FORGE_HOME (default process.cwd()), looking for memory.json, .forge-temp/memory_current.json or data/memory_current.json.

Details

Recent updates

README documents the data contract: forge-memory reads memory.json, .forge-temp/memory_current.json or data/memory_current.json under FORGE_HOME, with empty templates shipped in data/, and the tool list is published in the repo's plugin/tool table.

FAQ

How do I install forge-memory?
From a checkout: dsh plugin add ./plugins/forge-memory, or use the bundle tarball from the GitHub release. The npm form dsh-forge-memory does not currently resolve (404 verified 2026-09-11).
Where does it read memory from?
Under FORGE_HOME (default process.cwd()): memory.json, .forge-temp/memory_current.json or data/memory_current.json. Bring your own files — empty templates ship in data/.
How does recall work?
memory_recall uses BM25 to rank snippets for a query; memory_fold_list, memory_fold_preview and memory_fold_deep let the agent browse the folded index and expand only the sections it needs.

More plugins in Memory

Browse more in Memory

Guides for Memory plugins