melandlabs/opencontext
A temporal context graph, a memory API, retrieval primitives, and a multiple-platform integration mesh — designed to be embedded into any host process.
安装Install
dsh plugin --profile web add dsh-opencontext把 melandlabs/opencontext 加入你的 DSH 配置（web profile）即可启用。
READMEREADME
OpenContext
The agentic context runtime, powering applications that act on your behalf.
A temporal context graph, a memory API, retrieval primitives, and a multi-platform integration mesh — designed to be embedded into any host process or agents.
⭐ If you find opencontext useful, please consider giving us a star on GitHub! It helps more people discover the project and motivates us to keep building. 🙏
What is OpenContext?
OpenContext is the agentic context runtime that sits underneath an agentic application — and the substrate you build your own agent on top of. It is not a UI, a chat surface, or a model provider — it is the glue between the things that make an agent useful: durable memory, retrieval, context correction, multi-platform connectivity, scheduled awareness, and a deterministic loop engine, all behind one dependency.
→ Read docs/architecture.md for the full
data model, the lifecycle of a fact, and the transport surface map.
Who is it for?
OpenContext fits teams who need to engineer their context — that is, teams whose day-to-day work runs straight into the problems OpenContext was built to solve. Each bullet spells out the pain and how OpenContext addresses it:
- Software engineering teams. Decisions scatter across GitHub PRs, Linear tickets, Slack threads, and Notion docs — across people, tools, and quarters. New hires ask "why did we pick X?" and no one can answer. OpenContext's temporal graph stores every fact with
valid_from / valid_until, so "what did we believe last quarter?" is a real, citable query — not a guess. - Efficiency / productivity engineering teams. The people building internal automation for the rest of the company. They don't want another SaaS — they want a runtime they can drop into a CLI, an MCP server, or a daemon. OpenContext is library-first, and the deterministic Loop engine only invokes the LLM when there is real work, so it does not become a token-burning always-on loop.
- Office-assistant products. Assistants that live inside Telegram, iMessage, WhatsApp, Lark/Feishu, and friends. Same agent code, same context across channels.
IntegrationRecordhides credentials, rate-limits, and reconnect logic, whileplatform + messageIdis the natural audit trail for personal and work data. - Financial trading teams. Every order, rebalance, and risk decision needs to be traceable and auditable. The temporal graph plus append-only corrections mean "what was the strategy in April?" is a queryable fact, not a buried guess — and the trail lines up with MiFID II / SEC retention rules.
- Legal, healthcare and other audited domains. Law firms, hospitals, and similar teams where every judgement needs per-fact provenance, append-only corrections, and exportable compliance evidence.
- Multi-agent and autonomous-workflow authors. Need scheduled, deterministic wake-up instead of an LLM loop all the way down.
packages/loopships exactly that separation.
Features
| Capability | What it does | |
|---|---|---|
| 🧠 | Temporal Context Graph | A directed acyclic graph where every fact has valid_from / valid_until. Supersession, contradiction, and merge are first-class edges — corrections are append-only, not destructive. |
| 🔌 | Platform Integration Mesh | One uniform IntegrationRecord shape across Gmail, Slack, Telegram, Linear, Jira, iMessage, Feishu, Weixin, … — credential rotation, rate-limit handling, and reconnect logic live behind the adapter. |
| ⏰ | Deterministic Loop Engine | A scheduler that wakes up, decides whether there is real work, and only then calls into the agent runtime. LLM calls are not the foundation — they are the last step. |
| 🔍 | Retrieval Primitives | Chunking, embeddings, parsers (PDF/ZIP/text), sqlite-vec + pgvector + Chroma adapters. Mix backends without rewriting the recall pipeline. |
| 🤖 | Agent Runtime | AI SDK wrappers, sandbox providers (native / Claude / Vercel), MCP server, memory-consolidation job, image + audio generation. |
| 🪶 | Library-First API | Install once with pnpm add @melandlabs/opencontext and get the contracts, memory store, retrieval primitives, loop engine, and agent runtime. |
| 🛡️ | Audit + Encrypted Storage | Structured audit logging to ~/.opencontext/logs/audit.jsonl, Fernet symmetric encryption for secrets, URL allowlist/blocklist for outbound calls. |
Benchmarks
Third-party memory and long-context recall benchmarks (numbers current as of 2026-08):
| Benchmark | Score | What it measures |
|---|---|---|
| LongMemEval-S | 97.6% | Long-term memory recall across long sessions |
| LoCoMo-V2 | 97.4% | QA over long multimodal conversations |
| BEAM @ 10M | 67.0% | Factual recall at a 10M-token context window |
Quick Start
There are several ways to get opencontext into your project. Pick the one that matches what you're building.
1. Embed the runtime into your own app
pnpm add @melandlabs/opencontext
A 30-second example of the memory API:
import { createMemoryStore, getRawMessageManager } from "@melandlabs/opencontext";
// The store defaults to SQLite at MEMORY_STORE_DB_PATH (./memory.db by
// default). Each call returns an awaitable handle.
const store = await createMemoryStore();
const messages = await getRawMessageManager();
// A message is one fact: a single piece of content attributed to a user.
// `messageId` makes the call idempotent across re-ingest.
const now = Date.now();
await messages.storeMessages([
{
messageId: "msg-1",
userId: "u-42",
content: "User prefers dark mode in all tools",
platform: "test",
botId: "bot-1",
timestamp: now,
createdAt: now,
},
]);
// Unified search fans out to memory + insights + knowledge. Sources you
// haven't wired up just emit a warning — fine for a single-backend deploy.
const hits = await store.search({
userId: "u-42",
query: "What does the user prefer?",
limit: 5,
});
// hits.count — number of results
// hits.sources — which sub-indexes were actually consulted
// hits.warnings — per-source degradation (e.g. missing embedder)
2. Manage memory from the CLI
Write and query memory straight from the command line. add writes a
single raw message to the active manager with no LLM roundtrip, and
search runs a unified read across memory, insights, and knowledge.
pnpm add -g @melandlabs/opencontext # puts the `opencontext` bin on PATH
# Write a fact (auto-fills messageId, platform="cli", timestamp=now)
opencontext add --text "Rust achieves memory safety without GC"
# Write with full provenance for later consolidation
opencontext add \
--text "Discussed Q4 roadmap with the team" \
--source "meeting://2026-08-20" --kind experience \
--tag topic=roadmap --tag team=eng
# Plain hybrid search (RRF across memory + insights + knowledge)
opencontext search --query "memory safety" --k 5
# Inspect what would have been sent to the LLM, no synthesis call
opencontext search --query "what did we chat about last weekend" --context-only
# Script-friendly JSON
opencontext search --query "x" --json | jq '.results[].id'
add accepts --user (default "default"), --bot (default "default"),
--platform, --channel, --person, --source, --kind, --at,
and repeatable --tag key=value. search accepts --mode {auto|lex|sem},
--k, --threshold, repeatable --bot / --kind, --since / --until,
and --explain to surface reasoning + warnings alongside the hits.
Run opencontext <command> --help for the full flag list. See the
Getting Started tutorial
for the full flag reference and worked examples.
3. Run the HTTP daemon from npm
pnpm add -g @melandlabs/opencontext # puts the `opencontext` bin on PATH
opencontext http \
--embedding-provider local \
--memory-backend sqlite-vec \
--host 127.0.0.1 --port 7421
# Or, without a global install, via npx:
npx -y @melandlabs/opencontext http \
--embedding-provider local --memory-backend sqlite-vec
curl http://127.0.0.1:7421/health
4. Wire the MCP server into Claude Desktop / Cursor
opencontext mcp \
--embedding-provider local \
--memory-backend sqlite-vec
5. Use with DeepSeek Harness (DSH)
OpenContext is available as a DSH plugin that gives any DSH agent durable memory and retrieval-augmented context:
# Install the plugin from npm
dsh plugin --profile web add dsh-opencontext
# Confirm it's mounted
dsh --profile web --dump-config | grep dsh-opencontext
# ... should contain `id: dsh-opencontext`
# Start DSH web and verify
dsh web
# Visit http://127.0.0.1:3080/plugins and confirm dsh-opencontext shows "Enabled"
The plugin exposes 16 oc_* tools (e.g., oc_search, oc_remember, oc_memory_list) and automatically:
- Runs a recall waterfall on each turn to inject relevant historical context
- Captures user messages into durable memory
- Summarizes sessions at natural breakpoints (opt-in)
See plugins/dsh-opencontext/README.md for configuration options and the full tool reference.
6. Diagnose the install
opencontext doctor # human-readable health checks
opencontext doctor --json # CI-friendly { ok, exit, results } envelope
opencontext doctor --section memory-store
doctor is read-only and exits 0 on a healthy install. It scans nine
sections (runtime, filesystem, loop, memory-store, embedding,
policies, audit, security, integrations) and reports pass /
warn / fail for each. No auto-fix in v1.
Next: Tutorials — get started, user guide, developer guide, advanced patterns, and best practices
Examples
The examples/ workspace ships a runnable example per
capability area. Clone, install, and run:
git clone https://github.com/melandlabs/opencontext.git
cd opencontext/examples
pnpm install
pnpm test
See examples/README.md for the full walkthrough.
Why It Is Different
OpenContext is not a memory library and not a vector DB. It is a
runtime substrate — the @melandlabs/opencontext package bundles
contracts, memory-store, retrieval primitives, the loop engine, and
the agent runtime behind one dependency.
| Compared with… | opencontext adds |
|---|---|
| A flat vector DB (Pinecone, Weaviate, Qdrant) | A temporal graph — facts have valid_from / valid_until and get superseded, not just similarity-matched |
| A context/memory library | A runtime, not a library — HTTP daemon, MCP server, CLI, plus the integrations mesh and the loop engine |
| Wiring your own agent loop | A separable Loop engine that schedules when to wake the agent, instead of an LLM loop all the way down |
| Embedding opencontext just to get its integrations | Single-package install — one pnpm add gets every capability, no React/Next/Tauri required to use |
Architecture
┌────────────────────────────┐
│ Host application │ ← your UI, CLI, or daemon
│ (a reference app, │
│ or your own embedder) │
└─────────────┬──────────────┘
│
┌────────────────────────┴────────────────────────┐
│ @melandlabs/opencontext │
│ contracts · memory · rag · loop · agent │
└────────────────────────┬────────────────────────┘
│
┌─────────────────────────────┴─────────────────────────────┐
│ Storage backends │
│ sqlite-vec · postgres · indexeddb · chroma · pgvector │
└─────────────────────────────┬─────────────────────────────┘
│
┌─────────────────────────────┴─────────────────────────────┐
│ Integrations mesh (gmail, slack, …) │
└───────────────────────────────────────────────────────────┘
Full data-flow diagrams, transport surfaces, and storage backends are
in docs/architecture.md.
Documentation
Tutorials (Start Here)
docs/tutorials/README.md— Tutorial index and learning pathdocs/tutorials/00-getting-started.md— Get up and running in 5 minutesdocs/tutorials/01-user-guide.md— Understand the four verbs and temporal memorydocs/tutorials/02-developer-guide.md— Integrate OpenContext into your appdocs/tutorials/03-advanced-usage.md— Production patterns and advanced featuresdocs/tutorials/04-best-practices.md— Tips and common pitfallsdocs/tutorials/use-cases/README.md— Real-world use cases: personal assistant, support agent, research tracker
Architecture & Design
docs/architecture.md— data model, lifecycle, data plane and control planedocs/philosophy.md— why this shape- Each package's
README.md— API surface, examples, migration notes
Contributing
See CONTRIBUTING.md.
License
Apache-2.0. © 2026 Meland Labs.
相关推荐Related
- deepseek-ai/deepseek-harness DeepSeek Harness: Everything is a Plugin.
- deepseek-ai/deepseek-harness DeepSeek Harness (dsh) — 由 DeepSeek AI 开发的开源 agent harness（智能体框架）。一切皆插件：核心 + 插件 + 技能 = 可热插拔的 AI 编程工作台。DeepSeek Harness: …
-
ruvnet/ruflo
🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build c…
-
amruthpillai/reactive-resume
A one-of-a-kind resume builder that keeps your privacy in mind. Completely secure, customizable, portable, open-source a…
-
volcengine/OpenViking
Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
-
Molunerfinn/PicGo
:rocket: The Ultimate Image Uploader for Efficient Creators. Supports Obsidian, Typora, VS Code etc. and 60+ image hosti…
-
titanwings/colleague-skill
将冰冷的离别化为温暖的 Skill，欢迎加入数字生命1.0！Transforming cold farewells into warm skills? It's giving rebirth era. Welcome to Digital …
-
nocobase/nocobase
NocoBase is an open-source AI + no-code platform for building business systems fast. Instead of generating everything fr…