Friday

Self-hosted persistent cognitive memory layer for AI coding agents.

Persists architecture decisions, schemas, and constraints across sessions via the Model Context Protocol (MCP).

Overview

Quickstart

Python SDK

Client Setup (MCP)

Architecture

Benchmarks

API Reference

_{Friday Neural Studio — Real-time WebGL knowledge graph visualizer rendering service topologies, entity dependencies, and versioned facts.}

The Problem: Session Amnesia

Modern AI coding agents (Cursor, Claude Code, Antigravity, VS Code) excel at isolated code generation. However, in continuous engineering workflows, developers encounter a structural limitation: Session Amnesia.

Current workarounds fall into two flawed patterns:

                  ┌─────────────────────────────────────────────────────────┐
                  │          WHY STANDARD APPROACHES BREAK DOWN             │
                  └─────────────────────────────────────────────────────────┘

   1. Context Windows (RAM)           2. Static Rules Files             3. Standard Vector RAG
  ┌─────────────────────────┐       ┌─────────────────────────┐       ┌─────────────────────────┐
  │ • Ephemeral volatile    │       │ • Linear token tax      │       │ • Matches text phrasing,│
  │   memory (clears on     │       │   (2,500 tokens burned  │       │   NOT system topology   │
  │   every new thread)     │       │   on every trivial fix) │       │ • Blind to directed     │
  │ • Lost-in-the-middle    │       │ • Stale rules accumu-   │       │   call graphs & schema  │
  │   degradation on 50k+   │       │   late & conflict       │       │   dependencies          │
  │   token prompts         │       │ • Zero cross-tool sync  │       │ • Hallucinates blast    │
  │ • High latency & cost   │       │   (Cursor ≠ Claude CLI) │       │   radii of refactors    │
  └─────────────────────────┘       └─────────────────────────┘       └─────────────────────────┘

1. Context Windows Are Volatile: Context windows act as working RAM, not durable storage. Clearing a thread or restarting an agent resets state. Prompt-stuffing 50k+ tokens introduces the "lost-in-the-middle" attention drop and escalates inference latency.
2. Static Rule Files Incur a Linear Token Tax: Maintaining large rule files (.cursorrules, AGENTS.md) forces the model to re-read thousands of lines on every keystroke, leading to contradictory instructions and cross-editor fragmentation.
3. Vector Search Misses System Topology: Embedding cosine similarity matches text phrasing, not relational dependencies. Vector search cannot traverse directed graphs: $$\text{Table: accounts} \longrightarrow \text{FK: subscriptions} \longrightarrow \text{Service: BillingService} \longrightarrow \text{Worker: InvoicePoller}$$

Architecture: Multi-Layer Cognitive Substrate

Friday runs as a self-hosted background service providing a structured, four-tier memory substrate accessed via the Model Context Protocol (MCP):

┌────────────────────────────────────────────────────────────────────────────────────────┐
│               AI CODING CLIENTS (Cursor / Claude Code / Antigravity / VS Code)         │
└───────────────────────────────────────────┬────────────────────────────────────────────┘
                                            │
                                4 MCP Tools (stdio / HTTP)
                                ├── add_memory       (persist decisions & rationale)
                                ├── add_fact         (versioned immutable truths)
                                ├── memory_search    (targeted semantic recall)
                                └── get_context      (compiled multi-layer prompt)
                                            │
                                            ▼
┌────────────────────────────────────────────────────────────────────────────────────────┐
│                                 FRIDAY COGNITIVE ENGINE                                │
│                                                                                        │
│   Layer 1: Facts Ledger         Layer 2: Episodic Memory       Layer 3: Graph Topology │
│  ┌─────────────────────────┐   ┌───────────────────────────┐  ┌──────────────────────┐ │
│  │ Versioned SQLite        │   │ Mem0 + ChromaDB           │  │ Neo4j Property Graph │ │
│  │ • Deterministic truths  │   │ • Semantic decisions      │  │ • Directed call-trees│ │
│  │ • Conflict detection    │   │ • Vector similarity       │  │ • Schema blast-radius│ │
│  │ • Zero prompt overhead  │   │ • Sub-100ms retrieval     │  │ • Entity dependencies│ │
│  └─────────────────────────┘   └───────────────────────────┘  └──────────────────────┘ │
│                                                                                        │
│   • Auto-Graph Pipeline: LLM extraction wires entities into Neo4j automatically.       │
│   • Neural Studio: WebGL-based 3D graph visualizer for human and agent state auditing. │
│   • Persona Synchronization: /export/persona compiles canonical rules on-demand.       │
└────────────────────────────────────────────────────────────────────────────────────────┘

Architectural Comparison Matrix

DeepEval Benchmarks

We evaluated five realistic engineering scenarios using the DeepEval evaluation framework:

1. Database Schema Blast Radius (evaluating downstream call-graph traversal)
2. Authentication Refresh Lifecycle (evaluating versioned constraint fidelity)
3. Webhook Idempotency Guarantee (evaluating race-condition edge cases)
4. Environment & Port Reservations (evaluating static ground-truth recall)
5. Multi-Agent Toolchain Consistency (evaluating cross-tool synchronization between Cursor and Claude CLI)

Reproducing Benchmarks Locally

python benchmarks/benchmark_deepeval.py

Quickstart

Option A: One-Command Installation (Recommended)

Run the automated installer to check dependencies, generate configuration keys, and boot the stack:

curl -fsSL https://raw.githubusercontent.com/friday-memory/friday/main/install.sh | bash

Option B: Manual Setup via Docker Compose

1. Clone the repository

git clone https://github.com/friday-memory/friday.git
cd friday
cp .env.example .env

2. Configure environment (.env)

# Master API key for endpoint security
FRIDAY_API_KEY=choose_a_strong_password

# LLM provider for automated graph extraction (DeepSeek or Groq)
DEEPSEEK_API_KEY=your_api_key_here
DEEPSEEK_BASE_URL=https://api.deepseek.com

# Mem0 key for vector memory
MEM0_API_KEY=your_mem0_key_here

# Neo4j database credentials
NEO4J_PASSWORD=choose_a_secure_db_password

3. Launch services

make docker-up
# or: docker compose up -d

Services initialized:

• Friday Gateway API: http://localhost:80 (or http://localhost:8000)
• Neo4j Browser: http://localhost:7474
• Neural Studio UI: http://localhost/

4. Verify health

curl http://localhost/health

5. Persist initial context

curl -X POST http://localhost/add \
  -H "X-Brain-Key: your_strong_password" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Authentication uses JWT access tokens (15m expiration) with httpOnly refresh cookies. Implementation in gateway/auth.py.",
    "project": "CoreApp"
  }'

Python SDK (friday-memory)

Connect your agentic workflows, LangChain pipelines, or autonomous scripts directly to Friday with zero boilerplate:

pip install friday-memory

Synchronous Client

from friday import Friday

# Automatically resolves FRIDAY_URL and FRIDAY_API_KEY from environment
with Friday(api_key="your_secret_key", base_url="http://localhost:8000") as client:
    # 1. Health check
    status = client.health()
    print("Friday Status:", status["status"])

    # 2. Store architectural decision
    client.add_memory(
        "PostgreSQL 16 selected with pgvector for hybrid retrieval",
        project="backend-api",
    )

    # 3. Commit immutable ground-truth fact
    client.add_fact("Production database endpoint is db.internal.net:5432")

    # 4. Multi-layer search (L2 Facts + L3 ChromaDB + L4 Knowledge Graph)
    context = client.search("database connection configuration", project="backend-api")
    print(context["results"])

Asynchronous Client (FastAPI / Agent Workers)

import asyncio
from friday import AsyncFriday

async def main():
    async with AsyncFriday(api_key="your_secret_key") as client:
        # Commit context concurrently
        await client.add_memory("Redis cluster deployed for token bucket rate limiting")
        facts = await client.get_facts()
        print(f"Verified facts count: {len(facts)}")

asyncio.run(main())

LangChain Integration (FridayRetriever)

pip install "friday-memory[langchain]"

from friday.integrations.langchain import FridayRetriever
from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI

retriever = FridayRetriever(
    api_key="your_secret_key",
    base_url="http://localhost:8000",
    project="reeldm",
)

# Connect directly to LCEL chains
prompt = ChatPromptTemplate.from_template("""Answer using verified system memory:
{context}

Question: {question}""")

chain = {"context": retriever, "question": RunnablePassthrough()} | prompt | ChatOpenAI()

Client Setup (MCP)

Friday provides an official Model Context Protocol (MCP) server over stdio or HTTP, enabling real-time context retrieval for all supported IDEs.

 ┌───────────────────────┐
 │   Cursor (Desktop)    │──┐
 └───────────────────────┘  │
 ┌───────────────────────┐  │
 │    Claude Code CLI    │──┼── MCP Protocol (stdio transport)
 └───────────────────────┘  │   FRIDAY_URL="http://127.0.0.1:8000"
 ┌───────────────────────┐  │   BRAIN_API_KEY="your_secret_key"
 │    Antigravity IDE    │──┤
 └───────────────────────┘  │
 ┌───────────────────────┐  │
 │  Windsurf / VS Code   │──┘
 └───────────────────────┘
                            ▼
             ┌──────────────────────────────┐
             │     FRIDAY CENTRAL BRAIN     │
             │   (Localhost or Remote VM)   │
             │   FastAPI + Mem0 + Neo4j     │
             └──────────────────────────────┘

{
  "mcpServers": {
    "friday": {
      "command": "python",
      "args": ["-m", "mcp.server"],
      "cwd": "/path/to/friday",
      "env": {
        "FRIDAY_URL": "http://localhost:8000",
        "BRAIN_API_KEY": "your_secret_key"
      }
    }
  }
}

claude mcp add friday \
  -e FRIDAY_URL="http://localhost:8000" \
  -e BRAIN_API_KEY="your_secret_key" \
  -- python -m mcp.server

{
  "mcpServers": {
    "friday": {
      "command": "python",
      "args": ["-m", "mcp.server"],
      "cwd": "/path/to/friday",
      "env": {
        "FRIDAY_URL": "http://localhost:8000",
        "BRAIN_API_KEY": "your_secret_key"
      }
    }
  }
}

{
  "cline.mcpServers": {
    "friday": {
      "command": "python",
      "args": ["-m", "mcp.server"],
      "cwd": "/path/to/friday",
      "env": {
        "FRIDAY_URL": "http://localhost:8000",
        "BRAIN_API_KEY": "your_secret_key"
      }
    }
  }
}

Tool Reference

Connected agents automatically access four core MCP primitives:

Dynamic Directives Export (/export/persona)

Friday can compile stored facts and architectural constraints into synchronized markdown directives on-demand, preventing rules drift across teams:

# Export canonical AGENTS.md
curl -s "http://localhost/export/persona?target=agents" \
  -H "X-Brain-Key: your_key" > AGENTS.md

# Export Cursor .cursorrules
curl -s "http://localhost/export/persona?target=cursor" \
  -H "X-Brain-Key: your_key" > .cursorrules

Features

1. Automated Knowledge Graph Extraction

Every memory written via add_memory is analyzed asynchronously. Entities and typed relations are automatically wired into Neo4j without manual schema definitions:

Input:
"Billing engine connects to Stripe API for recurring charges. Webhook dispatched to /api/webhooks/stripe."

Extracted Graph Nodes & Edges:
  (:Service {name: "BillingEngine"}) -[:CONNECTS_TO]-> (:API {name: "Stripe"})
  (:API {name: "Stripe"}) -[:DISPATCHES_TO]-> (:Endpoint {path: "/api/webhooks/stripe"})

2. Neural Studio (3D Topology Visualizer)

A browser-based WebGL graph explorer (Three.js) for auditing agent memory:

• Cluster Topologies: Visualizes architectural components as a 3D force-directed graph.
• Entity Inspector: Inspect node connections, versioned facts, and raw vector chunks.
• Live CRUD: Create, rename, or link entities directly within the visual interface.
• High-Resolution Export: Export topology diagrams for technical documentation.

3. Versioned Facts Ledger

Deterministic project constants are recorded with immutable version history. Outdated statements are superseded rather than overwritten, preserving an audit trail:

# Add initial constraint
POST /facts -> {"content": "PostgreSQL 16 running on port 5432"}
# Recorded: id="c41b8a9", superseded=false

# Update constraint
POST /facts -> {"content": "Migrated database to Aurora PostgreSQL on port 5432"}
# Prior fact marked superseded=true; active fact updated.

Repository Structure

friday/
├── gateway/                 # FastAPI REST application & routing
├── layers/                  # Pluggable storage adapters (SQLite, ChromaDB, Neo4j)
├── pipelines/               # Background entity extraction & fact pipelines
├── orchestrator/            # Multi-layer retrieval router
├── mcp/                     # Model Context Protocol stdio server
├── studio/                  # Three.js Neural Studio visualizer
├── benchmarks/              # DeepEval evaluation suite
├── tests/                   # Pytest test suite
├── docker-compose.yml       # Production container definition
├── Makefile                 # Developer task automation
└── pyproject.toml           # Tooling & packaging configuration

API Reference

All authenticated endpoints require the X-Brain-Key request header.

Development

# Install dependencies
make install

# Run test suite
make test

# Code formatting & linting
make lint
make format

# Start local dev server
make dev

Contributing

Review CONTRIBUTING.md for pull request guidelines, commit conventions, and architectural standards.

License

Friday is licensed under the MIT License.