OpenContext - Give Your AI Assistant a Persistent Memory for Multi-Agent Workflows

🤔 Curiosity: How Can We Give AI Agents Persistent Memory?

When building AI-powered applications, one recurring frustration stands out: context gets lost. Every new chat, every session restart, every repo switch — the AI forgets everything. You end up re-explaining your project background, repeating architectural decisions, and sometimes the assistant makes wrong assumptions because it lacks historical context.

Curiosity: What if AI agents could remember your project decisions, technical constraints, and domain knowledge across sessions, repos, and tools? How would that transform the multi-agent development workflow?

Core Question: Can we create a unified context store that works across Cursor, Claude Code, Codex, and other AI coding assistants?

---

📚 Retrieve: Understanding OpenContext

The Problem

Before OpenContext	After OpenContext
📂 Hard to share context across repos/sessions	✅ Global context library works across all projects
🤷 Your ideas can’t be quickly perceived by Agent	✅ Agent loads your background & decisions automatically
🔒 Existing knowledge can’t be operated by Coding Agent	✅ Agent can directly read/write your knowledge base

What is OpenContext?

OpenContext is a lightweight personal context/knowledge store for AI assistants and coding tools. It enables the paradigm:

“Load history first, then act; ship, then persist.”

The tool provides:

• oc CLI — manage a global contexts/ library (folders/docs, manifests, search)
• MCP Server — so Cursor/Claude Code/Codex/Agents can call OpenContext as tools
• Skills + Slash Commands — user-level skills for Cursor/Claude Code/Codex
• Desktop App — manage/search/edit contexts with a native UI
• Web UI — browse/edit contexts locally (no install required)

Multi-Agent Integration Architecture

graph TB
    subgraph OpenContext["OpenContext Multi-Agent System"]
        A[User Context] --> B[OpenContext Store]

        B --> C[Cursor Agent]
        B --> D[Claude Code Agent]
        B --> E[Codex Agent]
        B --> F[Custom MCP Agents]

        C --> C1[opencontext-context]
        C --> C2[opencontext-search]
        C --> C3[opencontext-create]
        C --> C4[opencontext-iterate]

        D --> D1[Context Loading]
        D --> D2[Knowledge Search]
        D --> D3[Doc Creation]
        D --> D4[Insight Persistence]

        E --> E1[Skill-based Access]
        E --> E2[Tool Integration]

        F --> F1[MCP Protocol]
        F --> F2[Custom Tools]

        C1 --> G[Shared Knowledge Base]
        C2 --> G
        D1 --> G
        D2 --> G
        E1 --> G
        F1 --> G

        G --> H[Persistent AI Memory]
    end

    style B fill:#ff6b6b,stroke:#c92a2a,stroke-width:3px,color:#fff
    style G fill:#4ecdc4,stroke:#0a9396,stroke-width:2px,color:#fff
    style H fill:#ffe66d,stroke:#f4a261,stroke-width:2px,color:#000

---

💡 Innovation: Setting Up Multi-Agent Context Management

Quick Start (30 Seconds)

# 1. Install CLI
npm install -g @aicontextlab/cli

# 2. Initialize (prompts for tool setup; defaults to all)
cd your-project
oc init

# 3. Use slash commands (Cursor + Claude Code)
# /opencontext-context — load background before working
# /opencontext-search — find relevant docs
# /opencontext-create — create a new doc
# /opencontext-iterate — persist what you learned

Installed Components by oc init

Component	Location	Purpose
Cursor Commands	~/.cursor/commands	Slash commands for context operations
Claude Code Commands	~/.claude/commands	Slash commands integration
Cursor Skills	~/.cursor/skills/opencontext-*/SKILL.md	User-level skill definitions
Claude Code Skills	~/.claude/skills/opencontext-*/SKILL.md	Skill-based context access
Codex Skills	~/.codex/skills/opencontext-*/SKILL.md	Codex skill integration
MCP Config	~/.cursor/mcp.json, ~/.claude/mcp.json	MCP server configuration

CLI Commands Reference

Command	What it does
oc init	Initialize OpenContext + user-level tool integrations
oc folder ls	List folders
oc folder create <path> -d "desc"	Create a folder
oc doc create <folder> <name>.md -d "desc"	Create a document
oc doc ls <folder>	List documents
oc context manifest <folder>	Generate file list for AI to read
oc search "query"	Search documents
oc mcp	Start MCP server for MCP clients
oc ui	Start local Web UI

---

🛠️ Integrating with Multi-Agent Workflows

Workflow Pattern: Context-First Development

flowchart LR
    subgraph BeforeWork["Before Work"]
        A1[opencontext-context]
        A2[Load project background]
    end

    subgraph DuringWork["During Work"]
        B1[opencontext-search]
        B2[Search existing conclusions]
    end

    subgraph AfterWork["After Work"]
        C1[opencontext-iterate]
        C2[Record decisions]
    end

    A1 --> A2
    A2 --> B1
    B1 --> B2
    B2 --> C1
    C1 --> C2
    C2 -.-> A1

    style A1 fill:#ff6b6b,stroke:#c92a2a,stroke-width:2px,color:#fff
    style B1 fill:#4ecdc4,stroke:#0a9396,stroke-width:2px,color:#fff
    style C1 fill:#ffe66d,stroke:#f4a261,stroke-width:2px,color:#000

Slash Commands Deep Dive

Command	Description	Best For
/opencontext-help	Not sure which to use? Start here	New users
/opencontext-context	(Safe default) Load background before working	Starting any task
/opencontext-search	Discover existing docs (no auto index build)	Finding prior decisions
/opencontext-create	Draft a new doc/idea	Capturing new knowledge
/opencontext-iterate	Persist conclusions & citations	Ending sessions

Pro tip: Persist “acceptance criteria / common pitfalls / API contracts / dependency versions” as docs — highest ROI.

---

📊 Desktop App Features

For visual users who prefer a native UI:

Key Capabilities

1. Create folders and docs — Organize by project or topic
2. Search with Cmd/Ctrl + K — Searches titles, descriptions, and content
3. Semantic search — Configure API Key in Settings → Embedding (auto-indexed)
4. Cite to agents — Copy citations directly to Cursor/Claude/Agent

Citation Options

Action	Steps	Effect
Cite text snippet	Select text → Right-click → “Copy Citation”	Agent reads snippet + source
Cite document	Click citation icon next to doc title	Agent gets full doc + stable_id
Cite folder	Right-click folder → “Copy Folder Citation”	Agent batch-reads all docs inside

---

🔧 Search Configuration

Search Modes

Mode	Flag	Requirements
Keyword	--mode keyword	No embeddings needed
Vector	--mode vector	Requires embeddings + index
Hybrid	--mode hybrid	Requires embeddings + index (default)

Enabling Semantic Search (CLI)

# 1. Set API Key
oc config set EMBEDDING_API_KEY "<your_key>"

# 2. (Optional) Set base URL
oc config set EMBEDDING_API_BASE "https://api.openai.com/v1"

# 3. (Optional) Set Model
oc config set EMBEDDING_MODEL "text-embedding-3-small"

# 4. Build Index
oc index build

Warning: The oc index build command may use paid APIs. AI assistants are NOT allowed to auto-run it by default.

---

🤖 MCP Server Integration

OpenContext runs as a standard MCP (Model Context Protocol) server using stdio:

# Start manually
oc mcp

# User-level config locations:
# - Cursor: ~/.cursor/mcp.json
# - Claude Code: ~/.claude/mcp.json
# - Codex: ~/.codex/mcp.json

Example mcp.json Configuration

{
  "mcpServers": {
    "opencontext": {
      "command": "npx",
      "args": ["@aicontextlab/cli", "mcp"]
    }
  }
}

---

📈 Performance Comparison: With vs Without Context Store

Metric	Without OpenContext	With OpenContext	Improvement
Context Re-explanation	Every session	Once	⬇️ 95%
Decision Consistency	Variable	Tracked	⬆️ High
Cross-repo Knowledge	Manual	Automatic	⬆️ Seamless
Onboarding Time	Hours per project	Minutes	⬇️ 80%
Agent Accuracy	Context-limited	Context-enriched	⬆️ Significant

---

🎯 Best Practices for Multi-Agent Context

What to Persist

Document Type	Example Content	ROI
Acceptance Criteria	Feature requirements, success metrics	⭐⭐⭐
Common Pitfalls	Known bugs, edge cases, gotchas	⭐⭐⭐
API Contracts	Endpoint specs, data formats	⭐⭐
Dependency Versions	Lock file summaries, upgrade notes	⭐⭐
Architecture Decisions	ADRs, design rationale	⭐⭐
Domain Knowledge	Business logic, terminology	⭐⭐⭐

Folder Organization Example

~/.opencontext/contexts/
├── personal/
│   ├── preferences.md
│   └── coding-style.md
├── project-frontend/
│   ├── architecture.md
│   ├── component-patterns.md
│   └── known-issues.md
├── project-backend/
│   ├── api-design.md
│   ├── database-schema.md
│   └── deployment-notes.md
└── shared/
    ├── team-conventions.md
    └── dependency-policies.md

---

🚀 Environment Configuration

Default Paths

Purpose	Default Location
Contexts	~/.opencontext/contexts
Database	~/.opencontext/opencontext.db

Override with Environment Variables

export OPENCONTEXT_CONTEXTS_ROOT="/path/to/contexts"
export OPENCONTEXT_DB_PATH="/path/to/opencontext.db"

---

🤔 New Questions This Raises

1. Automatic Context Extraction: Can agents automatically identify and persist important decisions from conversations?
2. Context Versioning: How do we handle context evolution and maintain historical snapshots?
3. Team Collaboration: Can multiple developers share and sync context across a team?
4. Domain-Specific Templates: What context structures work best for different project types (web app, mobile, ML, etc.)?
5. Context Quality: How do we measure and improve the quality of persisted context?

Next Experiment: Building automated context extraction from conversation history using multi-agent orchestration.

---

References

Official Resources:

• OpenContext GitHub Repository
• OpenContext Usage Guide
• OpenContext Website
• NPM Package: @aicontextlab/cli

Related Tools & Frameworks:

• Model Context Protocol (MCP)
• Cursor IDE
• Claude Code
• OpenAI Codex

Multi-Agent Systems:

• LangGraph - Multi-Agent Workflows
• AutoGen - Multi-Agent Framework
• CrewAI - Multi-Agent Framework

Context Management Research:

• RAG: Retrieval-Augmented Generation
• Long-term Memory for LLMs
• Building Effective AI Agents (Anthropic)

Production Resources:

• LLM Engineering Best Practices
• Prompt Engineering Guide
• AI Agent Design Patterns