cognex 0.1.4

Persistent memory layer for AI agents — survives sessions, learns patterns, enables continuity

Cognex

Persistent memory layer for AI agents — survives sessions, learns patterns, enables continuity.

Give your AI coding assistant long-term memory, decision tracking, and trust management. Benchmarked to reduce context tokens by ~70% compared to manual context pasting.

---

Why Use It?

You: "Remember I prefer pytest over unittest"

Next session:
AI: "Got it — I'll use pytest for your tests instead."

Your AI forgets everything between sessions. Cognex fixes that.

Features

Feature	What It Does
Persistent Memory	Remembers preferences, facts, patterns across sessions
Decision Ledger	Tracks choices made and their outcomes
Trust Engine	Learns which tools you approve vs deny
Teleportation	Export your brain, load it on another machine
Swarm Mode	Turn natural language into multi-agent plans
MCP Prompts	5 built-in prompts for common workflows
Auto-Installer	One command setup for 7+ AI platforms

---

Quick Install

Option A — Auto-installer (Recommended)

# Install and auto-configure for your AI tool
uvx cognex --install

This detects your AI tool (Claude Code, OpenCode, Cursor, VS Code, etc.) and writes the correct config automatically.

Option B — One command (Terminal)

Mac/Linux:

curl -fsSL https://raw.githubusercontent.com/Gaurav7974/cognex/main/install.sh | sh

Windows (PowerShell):

irm https://raw.githubusercontent.com/Gaurav7974/cognex/main/install.ps1 | iex

These scripts automatically:

• Install cognex via pip
• Detect your AI tool (Claude Code, OpenCode, Cursor, Codex)
• Write the correct config file
• Print confirmation when done

---

Option C — Paste this prompt into your AI agent

If you're already inside Claude Code, OpenCode, Cursor or any MCP-compatible tool, just paste this prompt and your AI will configure everything:

Install and configure the cognex MCP server on my system.

Steps:
1. Run: pip install cognex
2. Detect which AI tool I am using (Claude Code / OpenCode / Cursor / Codex / VS Code)
3. Find the correct MCP config file for this tool
4. Add cognex to the config:
   - command: "cognex"
   - no extra args needed
5. Tell me to restart my AI tool
6. After I restart, call substrate_start_session with session_id "setup-verify" to confirm it works

Copy that prompt → paste into your AI tool → it handles the rest.

---

Option D — Manual (full control)

See the Configuration section below for step-by-step setup per tool.

---

Supported AI Tools

Tool	Works?
Claude Code	✅
Claude Desktop	✅
OpenCode	✅
Cursor	✅
Codex	✅
Any MCP-compatible tool	✅

---

Installation (Choose One)

Option 1: uvx (Recommended — no install needed)

uvx cognex

Option 2: pipx (isolated environment)

pipx install cognex

Option 3: pip (system-wide install)

pip install cognex

Option 4: Install from source (development)

git clone https://github.com/Gaurav7974/cognex
cd cognex
pip install -e .

Verify Installation

After any install method above:

cognex --help
# Should show: usage: cognex [-h] [--db-path ...] [--project ...] [--name ...] [--debug] [--install]

---

Configuration

Claude Code

Config file: ~/.claude.json (global) or .mcp.json (project root)

Add via CLI (recommended):

claude mcp add cognex -- uvx cognex

Or manually add to ~/.claude.json:

{
  "mcpServers": {
    "cognex": {
      "command": "uvx",
      "args": ["cognex"]
    }
  }
}

Or add to .mcp.json in your project root (team-shared):

{
  "mcpServers": {
    "cognex": {
      "command": "uvx",
      "args": ["cognex"]
    }
  }
}

Note: .mcp.json in project root is version-controlled and shared with your team. ~/.claude.json is user-specific and works across all projects.

---

Claude Desktop

Config file: Windows: %APPDATA%\Claude\claude_desktop_config.json Mac: ~/Library/Application Support/Claude/claude_desktop_config.json Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "cognex": {
      "command": "uvx",
      "args": ["cognex"]
    }
  }
}

---

OpenCode

Config file: ~/.config/opencode/opencode.json (global) opencode.json in project root (project-specific)

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "cognex": {
      "type": "local",
      "command": ["uvx", "cognex"],
      "enabled": true
    }
  }
}

Note: OpenCode uses "mcp" key (not "mcpServers") and requires "type": "local" for stdio servers. Config files are merged, not replaced — safe to add to existing config.

---

Cursor

Config file: Windows: %USERPROFILE%\.cursor\mcp.json Mac/Linux: ~/.cursor/mcp.json

{
  "mcpServers": {
    "cognex": {
      "command": "uvx",
      "args": ["cognex"]
    }
  }
}

Note: Cursor caps at 40 tools per config. Cognex uses 18 tools, well within limit.

---

VS Code (GitHub Copilot Agent Mode)

Config file: .vscode/mcp.json in your workspace (team-shared) Or run: MCP: Open User Configuration from Command Palette (global)

Note: VS Code uses "servers" key not "mcpServers"

{
  "servers": {
    "cognex": {
      "command": "uvx",
      "args": ["cognex"]
    }
  }
}

---

Cline (VS Code Extension)

Config file: Managed via Cline UI Windows: %APPDATA%\Code\User\globalStorage\saoudrizwan.claude-dev\settings\cline_mcp_settings.json Mac: ~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json

Open Cline → click MCP Servers icon → Configure tab → Edit Config

Add under mcpServers:

{
  "mcpServers": {
    "cognex": {
      "command": "uvx",
      "args": ["cognex"],
      "disabled": false,
      "alwaysAllow": []
    }
  }
}

---

Kilo Code (VS Code Extension)

Same format as Cline. Open Kilo Code → MCP Servers → Configure.

{
  "mcpServers": {
    "cognex": {
      "command": "uvx",
      "args": ["cognex"],
      "disabled": false
    }
  }
}

---

Windsurf

Config file: ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "cognex": {
      "command": "uvx",
      "args": ["cognex"]
    }
  }
}

---

Zed

Config file: ~/.config/zed/settings.json

Add under context_servers key:

{
  "context_servers": {
    "cognex": {
      "command": {
        "path": "uvx",
        "args": ["cognex"]
      }
    }
  }
}

---

Config format reference

Tool	Key	Config file
Claude Code	mcpServers	~/.claude.json or .mcp.json
Claude Desktop	mcpServers	claude_desktop_config.json
OpenCode	mcp	opencode.json
Cursor	mcpServers	~/.cursor/mcp.json
VS Code Copilot	servers	.vscode/mcp.json
Cline	mcpServers	cline_mcp_settings.json
Kilo Code	mcpServers	cline_mcp_settings.json
Windsurf	mcpServers	mcp_config.json
Zed	context_servers	settings.json

Key difference: OpenCode uses "mcp" and VS Code uses "servers". All others use "mcpServers".

---

After adding config

Completely close and reopen your AI tool. You should see 18 new cognex tools available.

To verify in Claude Code:

/mcp
→ Should show: cognex: connected

To verify in OpenCode:

Type /mcp in the chat
→ Should show cognex listed with Connected status

---

Restart Your AI Tool

After adding the config, completely close and reopen your AI tool. You should see 18 new tools available.

---

The 18 Tools

Session Management

Tool	Description
substrate_start_session	Start a new work session
substrate_end_session	End session with summary/metrics
substrate_process_transcript	Extract memories from conversation
substrate_report	Get memory health report

Memory

Tool	Description
memory_add	Add a memory (fact, preference, decision, pattern)
memory_search	Search memories with filters
memory_get_context	Get relevant context for current work
memory_decay	Age memories (auto-cleanup old ones)

Trust Engine

Tool	Description
trust_check	Check if tool needs approval
trust_record	Record approval/denial/violation
trust_get	Get trust score for a tool
trust_summary	Get trust overview

Decision Ledger

Tool	Description
ledger_record	Record a decision made
ledger_outcome	Record what happened after
ledger_find_similar	Find similar past decisions

Teleportation

Tool	Description
teleport_create_bundle	Export your brain to JSON
teleport_rehydrate	Import brain from another machine

Swarm

Tool	Description
swarm_compile_intent	Turn "build me an API" into a multi-agent plan

---

MCP Prompts

Cognex includes 5 built-in prompts accessible via MCP prompt protocol:

Prompt	Description
cognex://start-session	Initialize session with relevant memories
cognex://end-session	Generate session summary and save insights
cognex://context-for-task	Load context for a specific task
cognex://remember	Save important information to memory
cognex://weekly-summary	Get weekly activity and decision summary

Use these in any MCP-compatible client that supports prompts.

---

Example Usage

Remember a Preference

You: "I prefer using type hints everywhere"
AI: (calls memory_add)
→ Saved to your memory bank

Get Context Next Session

You: (start new session)
AI: (calls memory_get_context with "coding style")
→ Returns: "I prefer using type hints everywhere"
AI: "Got it — I'll add type hints throughout."

Track a Decision

You: "FastAPI or Flask?"
AI: "FastAPI has better type safety."
AI: (calls ledger_record)

Later...
You: "Did that work out?"
AI: (calls ledger_outcome with success: true)

Tool Trust Check

AI wants to run: rm -rf /
AI: (calls trust_check)
→ {requires_approval: true, trust_level: 0.2}
AI asks: "Can I delete everything?"

---

Where Data is Stored

All data stays local on your machine in SQLite:

Linux/Mac: ~/.cognex/cognex.db Windows: %USERPROFILE%\.cognex\cognex.db

Contains:

• memories — persistent memories
• sessions — session history
• trust_records — tool approval history
• decisions — decision ledger

---

Troubleshooting

"command not found" or "cognex" not recognized

Fix: Make sure you installed it:

pip install cognex

Or use uvx in your config:

"command": ["uvx", "cognex"]

Tools not appearing

1. Check the AI tool's developer console for errors
2. Try restarting the AI tool completely
3. Verify your JSON config is valid (use a JSON validator)

---

Development

git clone https://github.com/Gaurav7974/cognex
cd cognex
pip install -e ".[dev]"
pytest tests/ -v

---

File Structure

cognex/
├── .github/
│   ├── workflows/
│   │   ├── ci.yml
│   │   └── publish.yml
│   └── ISSUE_TEMPLATE/
│       ├── bug_report.md
│       └── feature_request.md
├── docs/
│   ├── configuration.md
│   ├── tools.md
│   └── examples.md
├── examples/
│   ├── basic_usage.py
│   ├── session_workflow.py
│   └── teleport_example.py
├── src/
│   ├── substrate/         ← Core memory system
│   └── substrate_mcp/     ← MCP server with 18 tools
├── tests/
│   ├── test_substrate.py
│   ├── test_layers.py
│   └── test_mcp_server.py
├── demo/
├── .gitignore
├── CHANGELOG.md
├── CONTRIBUTING.md
├── LICENSE
├── README.md
├── mcp.example.json       ← copy this to configure
└── pyproject.toml

---

License

MIT — free to use, modify, and distribute.

---

Need Help?

• Open an issue: https://github.com/Gaurav7974/cognex/issues
• PyPI page: https://pypi.org/project/cognex/
• Read the docs: see docs/ folder