melandlabs/opencontext

The context layer for AI agents: a temporal context graph with valid_from/valid_until facts, a platform integration mesh, and a deterministic loop engine that only calls the model when there is real work.

OpenContext is the agentic context runtime that sits underneath an agentic application: a temporal context graph where every fact carries valid_from / valid_until so supersession, contradiction and merge are first-class append-only edges; a platform integration mesh that normalizes Gmail, Slack, Telegram, Linear, Jira, iMessage, Feishu and Weixin behind one IntegrationRecord; a deterministic Loop engine that wakes on a schedule, decides whether there is real work, and only then calls the model; retrieval primitives (chunking, embeddings, PDF/ZIP parsers, sqlite-vec / pgvector / Chroma adapters); and an agent runtime with AI SDK wrappers, sandbox providers and an MCP server. Installed as a DSH plugin it gives any DSH agent durable memory and retrieval-augmented context through 16 oc_* tools such as oc_search, oc_remember and oc_memory_list. The README reports self-evaluated benchmark scores (LongMemEval-S 97.6%, LoCoMo-V2 97.4%, BEAM @ 10M 67.0%) produced by its own in-repo harnesses, and states these are not Agent Memory Leaderboard results.

Memory ★ 76
View on GitHub ↗

Install

dsh plugin --profile web add dsh-opencontext

Copied verbatim from the README's 'Use with DeepSeek Harness (DSH)' section, which also shows the verification steps dsh --profile web --dump-config | grep dsh-opencontext and a visit to http://127.0.0.1:3080/plugins. The library is also available as pnpm add @melandlabs/opencontext.

Compatibility

A DSH Web profile plugin. The README's DSH section says the plugin exposes 16 oc_* tools and runs a recall waterfall on each turn; the underlying runtime is library-first and Node-based.

Details

Alternatives

omdsh-dev/dsh-mnemon · csyangwen/dsh-memory-evolve · Co-Engram/Co-Engram

More plugins in Memory

Browse more in Memory

Guides for Memory plugins