Give your AI assistant persistent memory. Stop repeating explanations, and build smarter.
When to use this skill
When you need to keep context across sessions
When you need to record project background/decisions
When you need to search prior conclusions/lessons
When you need knowledge sharing in a Multi-Agent workflow
When you want to reduce repetitive background explanations
1. Core concepts
Problem
When working with an AI assistant, context gets lost (across sessions, repos, and dates). You end up repeating background, re-explaining decisions, and sometimes the assistant continues with incorrect assumptions.
Solution
OpenContext is a lightweight personal context/knowledge store for AI assistants.
npm install -g @aicontextlab/cli
# Or use npx
npx @aicontextlab/cli <command>
Initialize (run inside the repo)
cd your-project
oc init
Whatoc init does:
Prepare the global context store (on first run)
Generate user-level commands/skills + mcp.json for the selected tool
Update the repo's AGENTS.md
3. Slash Commands
Beginner-friendly commands
Command
Purpose
/opencontext-help
When you don't know where to start
/opencontext-context
(Recommended default) Load background before work
/opencontext-search
Search existing documents
/opencontext-create
Create a new document/idea
/opencontext-iterate
Store conclusions and citations
Install locations
# Slash Commands
Cursor: ~/.cursor/commands
Claude Code: ~/.claude/commands
# Skills
Cursor: ~/.cursor/skills/opencontext-*/SKILL.md
Claude Code: ~/.claude/skills/opencontext-*/SKILL.md
Codex: ~/.codex/skills/opencontext-*/SKILL.md
# MCP Config
Cursor: ~/.cursor/mcp.json
Claude Code: ~/.claude/mcp.json
4. Core CLI commands
Folder/document management
# List folders
oc folder ls --all
# Create folder
oc folder create project-a -d "My project"
# Create document
oc doc create project-a design.md -d "Design doc"
# List documents
oc doc ls project-a
Search & manifest
# Search (keyword/hybrid/vector)
oc search "your query" --mode keyword --format json
# Generate a manifest (list of files the AI should read)
oc context manifest project-a --limit 10
Search modes
Mode
Description
Requirements
--mode keyword
Keyword-based search
No embeddings required
--mode vector
Vector search
Embeddings + index required
--mode hybrid
Hybrid (default)
Embeddings + index required
Embedding configuration (for semantic search)
# Set API key
oc config set EMBEDDING_API_KEY "<<your_key>>"
# (Optional) Set base URL
oc config set EMBEDDING_API_BASE "https://api.openai.com/v1"
# (Optional) Set model
oc config set EMBEDDING_MODEL "text-embedding-3-small"
# Build index
oc index build
LM Studio (Local Embeddings — no API key required)
LM Studio exposes an OpenAI-compatible /v1/embeddings endpoint, so OpenContext can use it as a local embedding provider.
Step 1 — Load an embedding model in LM Studio
In the LM Studio app, go to Discover and download an embedding model:
Model
Size
Notes
nomic-ai/nomic-embed-text-v1.5-GGUF
~90 MB
General purpose, recommended
CompendiumLabs/bge-large-en-v1.5-gguf
~330 MB
Higher accuracy
second-state/All-MiniLM-L6-v2-Embedding-GGUF
~23 MB
Lightweight
Step 2 — Start the LM Studio server
lms server start
# Verify the embeddings endpoint is available
curl http://localhost:1234/v1/models
Step 3 — Configure OpenContext
# Point to LM Studio's local server
oc config set EMBEDDING_API_BASE "http://localhost:1234/v1"
# LM Studio requires any non-empty value for the key field
oc config set EMBEDDING_API_KEY "lm-studio"
# Set the embedding model ID (must match what is loaded in LM Studio)
oc config set EMBEDDING_MODEL "nomic-ai/nomic-embed-text-v1.5-GGUF"
# Build the index
oc index build
Verify
# Quick curl test
curl -s http://localhost:1234/v1/embeddings \
-H "Content-Type: application/json" \
-d '{"model":"nomic-ai/nomic-embed-text-v1.5-GGUF","input":"hello"}' \
| jq '.data[0].embedding | length'
# → should print a non-zero dimension (e.g. 768)
# Then test OpenContext search
oc search "your query" --mode vector
Note : LM Studio must be running whenever you run oc index build or use --mode vector / --mode hybrid search. --mode keyword works without a running server.
5. MCP Tools
OpenContext MCP Tools
oc_list_folders # List folders
oc_list_docs # List documents
oc_manifest # Generate manifest
oc_search # Search documents
oc_create_doc # Create document
oc_get_link # Generate stable link
Multi-Agent integration
# Gemini: large-scale analysis
ask-gemini "Analyze the structure of the entire codebase"
# Codex: run commands
shell "docker-compose up -d"
# OpenContext: store results
oc doc create project-a conclusions.md -d "Analysis conclusions"
6. Multi-Agent workflow integration
Orchestration Pattern
[Claude] Plan
↓
[Gemini] Analysis/research + OpenContext search
↓
[Claude] Write code
↓
[Codex] Run/test
↓
[Claude] Synthesize results + store in OpenContext
Practical example: API design + implementation + testing
# 1. [Claude] Design API spec using the skill
/opencontext-context # Load project background
# 2. [Gemini] Analyze a large codebase
ask-gemini "@src/ Analyze existing API patterns"
# 3. [Claude] Implement code based on the analysis
# (Use context loaded from OpenContext)
# 4. [Codex] Test and build
shell "npm test && npm run build"
# 5. [Claude] Create final report + store conclusions
/opencontext-iterate # Store decisions and lessons learned
7. Recommended daily workflow
Before work (1 min)
/opencontext-context
Load project background + known pitfalls
During work
/opencontext-search
Search existing conclusions when unsure
After work (2 min)
/opencontext-iterate
Record decisions, pitfalls, and next steps
High-ROI document types
Acceptance Criteria - acceptance criteria
Common Pitfalls - common pitfalls
API Contracts - API contracts
Dependency Versions - dependency versions
8. Stable links (Stable Links)
Keep links stable across renames/moves by referencing document IDs:
[label](oc://doc/<stable_id>)
Generate a link via CLI
oc doc link <doc_path>
Generate a link via MCP
oc_get_link doc_path="Product/api-spec"
9. Desktop App & Web UI
Desktop App (recommended)
Manage/search/edit context with a native UI
Use without the CLI
Automatic index builds (in the background)
Citation features:
Action
How
Result
Cite text snippet
Select text → right-click → "Copy Citation"
Agent reads the snippet + source
Cite document
Click the citation icon next to the document title
