GitHub - project-nomos/nomos: Autonomous AI agent with persistent memory, multi-agent teams, and 35+ skills. Deploy to Slack/Discord/Telegram/WhatsApp in minutes. Self-hosted, MIT licensed.

An autonomous AI agent that remembers everything, works everywhere, and never stops learning
Deploy to Slack, Discord, Telegram, WhatsApp — with persistent memory, multi-agent teams, and a web dashboard. All TypeScript, fully open source.

Quick Start · Features · Daemon · Channels · Skills · Contributing

What is Nomos?

Most AI assistants forget you the moment a conversation ends. Nomos doesn't.

Nomos is an autonomous AI agent platform that remembers every interaction, connects to your tools and channels, and gets smarter over time. It runs as a persistent background daemon — always on, always learning — reachable from Slack, Discord, Telegram, WhatsApp, the terminal, or any gRPC/WebSocket client.

It comes with persistent vector memory across sessions and channels, multi-agent team orchestration for complex tasks, smart model routing to optimize cost and quality, 27 bundled skills, a web management dashboard, and built-in image and video generation.

What powers Nomos? Under the hood, Nomos is built on Anthropic's Claude Agent SDK — inheriting the full agent loop, built-in tools, streaming, and compaction. Nomos adds the infrastructure layer on top: memory, channels, teams, scheduling, and a management UI.

Get Running in 2 Minutes

# Homebrew (recommended)
brew tap project-nomos/nomos https://github.com/project-nomos/nomos
brew install project-nomos/nomos/nomos

# Then just:
nomos chat

That's it. A browser-based setup wizard handles the rest — database connection, API provider, assistant personality, and channel integrations. Everything is saved encrypted in PostgreSQL.

Other installation methods

npm (GitHub Packages)

npm install -g @project-nomos/nomos --registry=https://npm.pkg.github.com

Docker Compose (includes database)

git clone https://github.com/project-nomos/nomos.git
cd nomos
cp .env.example .env
# Edit .env with your ANTHROPIC_API_KEY (or OPENROUTER_API_KEY)
docker compose up -d

The agent is accessible via gRPC on port 8766 and WebSocket on port 8765.

Docker (standalone)

docker run -d --name nomos \
  -e DATABASE_URL=postgresql://nomos:nomos@host.docker.internal:5432/nomos \
  -e ANTHROPIC_API_KEY=sk-ant-... \
  -p 8766:8766 -p 8765:8765 \
  ghcr.io/project-nomos/nomos:latest

From source

git clone https://github.com/project-nomos/nomos.git
cd nomos
pnpm install
pnpm build
pnpm link --global
nomos chat

Prerequisites

Node.js >= 22
PostgreSQL with the pgvector extension
One of: Anthropic API key, Google Cloud credentials (Vertex AI), OpenRouter API key, or a local Ollama instance
Google Cloud credentials for embeddings (optional — falls back to full-text search when unavailable)

What You Get

	Feature	What it does
🧠	Memory That Persists	Every conversation auto-indexed into pgvector. Recall anything from any session or channel.
💬	6 Channel Integrations	Slack, Discord, Telegram, WhatsApp, iMessage — thin adapters, one agent runtime.
👥	Multi-Agent Teams	Coordinator + parallel workers. Hand off complex tasks, get synthesized results.
⚡	Smart Model Routing	Route by complexity across any provider — cloud, local, or hybrid. Cut costs automatically.
🌐	Multiple API Providers	5 providers supported: direct API, Vertex AI, OpenRouter, Ollama, or custom.
🎨	Image & Video Gen	Gemini image + Veo video generation, conversational — just ask.
🖥️	Web Dashboard	Next.js 16 settings UI with setup wizard. No YAML editing.
🧩	27 Bundled Skills	Three-tier loading: bundled, personal, project. Create your own in minutes.
🔒	Secrets Encrypted at Rest	AES-256-GCM for all API keys and tokens. Auto-key on first run.
🧠	Adaptive Memory	Extracts facts, preferences, corrections. Builds a persistent user model.
🔄	Self-Improvement	Nomos can analyze its own code, implement fixes, and open PRs to itself.

Memory That Persists Across Sessions and Channels

Every conversation is automatically indexed into a PostgreSQL-backed vector store. When the agent needs context from a past interaction — even one that happened in a different channel weeks ago — it finds it.

Under the hood: pgvector with hybrid retrieval (vector cosine similarity + full-text search, fused via RRF). Embeddings via Vertex AI gemini-embedding-001 (768 dims). Falls back to FTS when embeddings aren't available.

Adaptive Memory & User Model

When enabled (NOMOS_ADAPTIVE_MEMORY=true), the agent extracts structured knowledge from every conversation — facts, preferences, and corrections — using a lightweight LLM call (Haiku by default). Extracted knowledge accumulates into a persistent user model that personalizes responses across sessions.

Knowledge extraction — facts about you, your projects, tech stack; preferences for coding style, communication, tools
Confidence-weighted — repeated confirmations increase confidence; contradictions decrease it
Prompt injection — high-confidence entries (>=0.6) are auto-injected into the system prompt

Self-Improvement — An Agent That Evolves Itself

Nomos has a built-in self-improve skill that lets it analyze its own codebase, implement changes, and open pull requests to itself — all autonomously. Ask it to fix a bug, add a feature, write tests, or refactor its own code.

How it works:

Clones a fresh copy of its own repo (never modifies the running instance)
Analyzes the codebase and implements the requested change
Runs all checks (pnpm check, pnpm test, pnpm build)
Opens a PR for your review

Just say "improve yourself", "add tests for the chunker", or "fix your session cleanup logic" — and review the PR when it's ready.

Connect From Anywhere

Protocol	Port	Purpose
gRPC	8766	Primary protocol for CLI, web, and mobile clients
WebSocket	8765	Legacy protocol (maintained for backward compatibility)
Ink CLI	—	Interactive terminal REPL with streaming markdown
Next.js Web UI	3456	Settings dashboard and management interface

Web-Based Management Dashboard

A full Next.js 16 app for onboarding, assistant configuration, channel management, and advanced settings — no YAML editing required. Includes a setup wizard (/setup), dashboard overview, per-channel integration management, and database/memory admin.

Extend Without Forking

Add capabilities without modifying core code — drop in a skill file, point to an MCP server, or change a config value:

27 bundled skills covering GitHub, Google Workspace, document generation, design, and more
Three-tier skill loading: ship your own skills per-project (./skills/), per-user (~/.nomos/skills/), or contribute upstream
MCP server support — connect any MCP-compatible tool server via .nomos/mcp.json
Config in the database — everything persists in PostgreSQL, env vars work as fallback

Deploy to 6 Platforms in Minutes

Platform	Mode	Transport
Slack	Bot Mode	Socket Mode
Slack	User Mode	Socket Mode + OAuth (multi-workspace, draft-before-send)
Discord	Bot	Gateway
Telegram	Bot	grammY
WhatsApp	Bridge	Baileys
iMessage	Read-only bridge	macOS Messages.app SQLite

Each adapter is a thin layer (~50-100 LOC). All agent logic is centralized in AgentRuntime; adapters just route messages in and responses out.

Smart Model Routing — Cut Costs Without Cutting Quality

Route queries to the right model automatically based on complexity. Works with any provider — Anthropic, OpenRouter, Ollama, or your own endpoint. Use Claude models, open-source local models, or mix and match:

Simple (greetings, short questions) -> fast, cheap model (e.g. claude-haiku-4-5, llama3, or any local model)
Moderate (general tasks) -> balanced model (e.g. claude-sonnet-4-6, mistral-large)
Complex (coding, reasoning, multi-step) -> most capable model (e.g. claude-opus-4-6, deepseek-r1)

Run fully local with Ollama, optimize cloud costs across tiers, or combine local models for simple tasks with cloud models for complex ones.

Enable with NOMOS_SMART_ROUTING=true.

Multiple API Providers

Run Nomos with the provider that fits your setup:

Anthropic — direct API access (default)
Vertex AI — Google Cloud managed API
OpenRouter — unified billing, usage tracking via openrouter.ai
Ollama — run local models via LiteLLM proxy
Custom — any Anthropic-compatible endpoint via ANTHROPIC_BASE_URL

Switch providers in the Settings UI or via NOMOS_API_PROVIDER. See API Providers for setup details.

Secrets Encrypted at Rest

API keys, OAuth tokens, and integration secrets never touch the database in plaintext. Everything is encrypted with AES-256-GCM before storage. A key is auto-generated on first run (~/.nomos/encryption.key), or bring your own via ENCRYPTION_KEY.

Multi-Agent Teams — Parallelize Complex Work

Hand off a big task and let multiple agents tackle it simultaneously. A coordinator breaks the problem down, spawns parallel workers, and synthesizes their results into one response.

User prompt --> Coordinator --+--> Worker 1 (subtask A) --+
                              +--> Worker 2 (subtask B) --+--> Coordinator --> Final response
                              +--> Worker 3 (subtask C) --+

Workers run in parallel with independent SDK sessions and scoped prompts
Failed workers are gracefully handled — other workers' results are preserved
Trigger with /team prefix: /team Research React vs Svelte and write a comparison
Configure via NOMOS_TEAM_MODE=true and NOMOS_MAX_TEAM_WORKERS=3

Image & Video Generation — Built In, Not Bolted On

Generate images (via Gemini) and videos (via Veo) directly from conversation. Ask your agent to create a logo, a product mockup, or a short video clip — it handles the API calls, saves the output, and returns the file. Opt-in via Settings UI or env vars.

Image generation — gemini-3-pro-image-preview. Photorealistic images, illustrations, logos, concept art.
Video generation — veo-3.0-generate-preview. Short cinematic clips with camera and style control.
Shared API key — both use GEMINI_API_KEY from Google AI Studio

Beyond the Basics

Digital marketing suite — Google Ads + Analytics via MCP, with team mode workflows and daily briefings
Slack User Mode — act as you: draft responses to your DMs for approval, then send as the authenticated user
Cron / scheduled tasks — run prompts on a schedule with configurable targets and delivery modes
Session persistence — all conversations stored in PostgreSQL with auto-resume and compaction
Personalization — user profile, agent identity, SOUL.md personality, per-agent configs
30+ slash commands — model switching, memory search, session management, and more
Proactive messaging — send outbound messages to any channel outside the reply flow
Pairing system — 8-character codes with TTL for channel access control

Daemon Mode

The daemon turns Nomos into an always-on, multi-channel AI gateway. It boots an agent runtime, gRPC + WebSocket servers, channel adapters, and a cron engine — then processes incoming messages from all sources through a per-session message queue.

How the Daemon Works

                         +-------------------+
                         |     Gateway       |
                         | (orchestrator)    |
                         +--------+----------+
                                  |
      +------------+--------------+--------------+----------+
      |            |              |              |          |
+-----v------+ +--v-----+ +-----v--------+ +---v------+ +-v----------+
| gRPC       | | WS     | | Channel      | | Cron     | | Draft      |
| Server     | | Server | | Manager      | | Engine   | | Manager    |
| (port 8766)| | (8765) | | (adapters)   | |(schedule)| | (Slack UM) |
+-----+------+ +---+----+ +-----+--------+ +---+------+ +--+---------+
      |             |            |              |            |
      +------+------+------+----+------+-------+------+-----+
                            |
                   +--------v---------+
                   |  Message Queue   |
                   |  (per-session    |
                   |   FIFO)          |
                   +--------+---------+
                            |
                   +--------v---------+
                   |  Agent Runtime   |
                   |  (Agent SDK)     |
                   +------------------+

Gateway boots all subsystems and installs signal handlers for graceful shutdown.
Channel adapters register automatically based on which environment variables are present (e.g., SLACK_BOT_TOKEN enables the Slack adapter).
Message queue serializes messages per session key — concurrent messages to the same conversation are processed sequentially; messages for different sessions process in parallel.
Agent runtime loads config, profile, identity, skills, and MCP servers once at startup, then processes each message through the agent SDK.

nomos daemon start    # Background mode
nomos daemon run      # Development mode (foreground with logs)
nomos daemon status   # Check if running
nomos daemon logs     # Tail logs

Channel Integrations

Each channel adapter is automatically registered when its required environment variables are present. For detailed setup guides, see docs/integrations/.

Slack

SLACK_BOT_TOKEN=xoxb-...          # Bot User OAuth Token
SLACK_APP_TOKEN=xapp-...          # App-Level Token (Socket Mode)
SLACK_ALLOWED_CHANNELS=C123,C456  # Optional: restrict to specific channels

Responds to @mentions and direct messages. Supports threaded conversations.

Slack User Mode — add SLACK_USER_TOKEN=xoxp-... to act as you rather than as a bot. Generates draft responses for your approval before sending. See docs/integrations/slack-user-mode.md.

Discord

DISCORD_BOT_TOKEN=...                     # Bot token from Discord Developer Portal
DISCORD_ALLOWED_CHANNELS=123456,789012    # Optional: restrict to specific channels

Telegram

TELEGRAM_BOT_TOKEN=...                    # Token from @BotFather
TELEGRAM_ALLOWED_CHATS=123456,-789012     # Optional: restrict to specific chats

WhatsApp

WHATSAPP_ENABLED=true
WHATSAPP_ALLOWED_CHATS=15551234567@s.whatsapp.net

Uses Baileys with QR code auth (no Meta Business API required). Auth state persisted to ~/.nomos/whatsapp-auth/.

Data Integrations (via MCP)

Integration	MCP Server	Guide
Google Ads	google-ads-mcp	Setup
Google Analytics	analytics-mcp	Setup
Google Workspace	@googleworkspace/cli	Setup

API Providers

Provider	Description	Guide
Anthropic (default)	Direct Anthropic API	Set `ANTHROPIC_API_KEY`
Vertex AI	Google Cloud Vertex AI	Set `CLAUDE_CODE_USE_VERTEX=1` + GCP credentials
OpenRouter	Anthropic models via OpenRouter	Setup guide
Ollama	Local models via LiteLLM proxy	Setup guide
Custom	Any Anthropic-compatible endpoint	Set `ANTHROPIC_BASE_URL`

Skills System

Skills are markdown files (SKILL.md) with YAML frontmatter that provide domain-specific instructions to the agent. Loaded from three tiers: bundled (skills/) -> personal (~/.nomos/skills/) -> project (./skills/).

All 33 bundled skills

Skill	Description
`algorithmic-art`	Generative art and creative coding
`apple-notes`	Apple Notes integration
`apple-reminders`	Apple Reminders integration
`brand-guidelines`	Brand and style guide creation
`canvas-design`	Canvas-based design generation
`digital-marketing`	Google Ads + Analytics campaigns, team mode analysis workflows
`discord`	Discord bot and integration help
`doc-coauthoring`	Collaborative document writing
`docx`	Word document generation
`frontend-design`	Frontend UI/UX design guidance
`github`	GitHub workflow and PR management
`image-generation`	Image generation with Gemini (prompts, styles, capabilities)
`gws-gmail`	Gmail API (messages, drafts, labels)
`gws-drive`	Google Drive (files, folders, permissions)
`gws-calendar`	Google Calendar (events, scheduling)
`gws-sheets`	Google Sheets (read, write, append)
`gws-docs`	Google Docs (create, read, edit)
`gws-slides`	Google Slides (presentations)
`gws-shared`	Google Workspace shared auth and conventions
`internal-comms`	Internal communications drafting
`mcp-builder`	MCP server development
`pdf`	PDF document generation
`pptx`	PowerPoint presentation generation
`self-improve`	Self-improvement and learning
`skill-creator`	Create new skills from prompts
`slack`	Slack app and integration help
`slack-gif-creator`	Slack GIF creation
`telegram`	Telegram bot and integration help
`theme-factory`	Theme and color scheme generation
`video-generation`	Video generation with Veo (camera, styles, prompt guidance)
`weather`	Weather information and forecasts
`web-artifacts-builder`	Web artifact (HTML/CSS/JS) creation
`webapp-testing`	Web application testing guidance
`whatsapp`	WhatsApp integration help
`xlsx`	Excel spreadsheet generation

Creating a custom skill

mkdir -p ~/.nomos/skills/my-skill
cat > ~/.nomos/skills/my-skill/SKILL.md << 'EOF'
---
name: my-skill
description: "What this skill does"
---

# My Skill

Instructions for the agent when this skill is active...
EOF

The bundled skill-creator skill can also generate skills on your behalf via conversation.

Development

pnpm dev                # Run in dev mode (tsx, no build needed)
pnpm build              # Build with tsdown -> dist/index.js
pnpm typecheck          # TypeScript type check (tsc --noEmit)
pnpm test               # Run tests (vitest)
pnpm lint               # Lint (oxlint)
pnpm check              # Full check (format + typecheck + lint)
pnpm daemon:dev         # Run daemon in dev mode (tsx)

See CONTRIBUTING.md for development setup, architecture, code conventions, and how to submit pull requests.

Configuration Reference

Configuration is loaded with the following precedence: Database > environment variables > hardcoded defaults. API keys and secrets are stored encrypted (AES-256-GCM) in the integrations table.

Required

Variable	Description	Default
`DATABASE_URL`	PostgreSQL connection string (must have pgvector)	--

Provider (set one)

Variable	Description	Default
`ANTHROPIC_API_KEY`	Anthropic direct API key	--
`CLAUDE_CODE_USE_VERTEX`	Set to `1` to use Vertex AI	--
`GOOGLE_CLOUD_PROJECT`	Google Cloud project ID (for Vertex AI)	--
`CLOUD_ML_REGION`	Vertex AI region	--

Model and behavior

Variable	Description	Default
`NOMOS_MODEL`	Default Claude model	`claude-sonnet-4-6`
`NOMOS_PERMISSION_MODE`	Tool permission mode (default, acceptEdits, plan, dontAsk, bypassPermissions)	`acceptEdits`
`NOMOS_SMART_ROUTING`	Enable complexity-based model routing	`false`
`NOMOS_TEAM_MODE`	Enable multi-agent team orchestration	`false`
`NOMOS_MAX_TEAM_WORKERS`	Max parallel workers in team mode	`3`
`NOMOS_ADAPTIVE_MEMORY`	Enable knowledge extraction and user model learning	`false`
`NOMOS_IMAGE_GENERATION`	Enable image generation via Gemini	`false`
`GEMINI_API_KEY`	Gemini API key (shared by image and video generation)	--
`NOMOS_VIDEO_GENERATION`	Enable video generation via Veo	`false`
`ANTHROPIC_BASE_URL`	Custom Anthropic API base URL (Ollama, LiteLLM, etc.)	--

Channel integrations

Variable	Description	Default
`SLACK_BOT_TOKEN`	Slack Bot User OAuth Token	--
`SLACK_APP_TOKEN`	Slack App-Level Token (Socket Mode)	--
`DISCORD_BOT_TOKEN`	Discord bot token	--
`TELEGRAM_BOT_TOKEN`	Telegram bot token from @BotFather	--
`WHATSAPP_ENABLED`	Set to `true` to enable WhatsApp	--

See .env.example for the complete list of all configuration options.

gRPC & WebSocket Protocols

gRPC

Clients communicate via gRPC on localhost:8766. The service is defined in proto/nomos.proto.

service NomosAgent {
  rpc Chat (ChatRequest) returns (stream AgentEvent);
  rpc Command (CommandRequest) returns (CommandResponse);
  rpc GetStatus (Empty) returns (StatusResponse);
  rpc ListSessions (Empty) returns (SessionList);
  rpc GetSession (SessionRequest) returns (SessionResponse);
  rpc ListDrafts (Empty) returns (DraftList);
  rpc ApproveDraft (DraftAction) returns (DraftResponse);
  rpc RejectDraft (DraftAction) returns (DraftResponse);
  rpc Ping (Empty) returns (PongResponse);
}

The .proto file can generate native clients for iOS (Swift), Android (Kotlin), and other platforms.

WebSocket (Legacy)

The WebSocket server runs on ws://localhost:8765 for backwards compatibility. See the protocol documentation for message formats.

Slash Commands

Command	Description
`/clear`	Clear conversation context
`/compact`	Compact conversation to reduce context usage
`/status`	Show system status overview
`/model <name>`	Switch model
`/thinking <level>`	Set thinking level (off, minimal, low, medium, high, max)
`/profile set <key> <value>`	Set profile field (name, timezone, workspace, instructions)
`/identity set <key> <value>`	Set agent identity (name, emoji)
`/skills`	List loaded skills
`/memory search <query>`	Search the vector memory
`/drafts`	List pending draft responses (Slack User Mode)
`/approve <id>`	Approve a draft
`/config set <key> <value>`	Change a setting
`/tools`	List available tools
`/mcp`	List MCP servers
`/quit`	Exit Nomos

Settings Web UI

A local Next.js app at settings/ for managing your assistant via the browser (port 3456).

Route	Description
`/setup`	5-step onboarding wizard (database, API, personality, channels, ready)
`/dashboard`	Overview: assistant status, model, active channels, memory stats, quick actions
`/settings`	Assistant identity, API config, model, advanced settings
`/integrations`	Channel overview and per-platform configuration
`/admin/database`	Database connection and migration status
`/admin/memory`	Memory store stats and management

nomos settings              # Start on http://localhost:3456 and open browser
nomos settings --port 4000  # Custom port

Contributing

Contributions are welcome. See CONTRIBUTING.md for development setup, testing, code conventions, and how to submit pull requests.

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 26 Commits
.claude		.claude
.github		.github
.husky		.husky
Formula		Formula
autonomous		autonomous
docs		docs
images		images
proto		proto
scripts		scripts
settings		settings
skills		skills
src		src
.dockerignore		.dockerignore
.env.example		.env.example
.gitignore		.gitignore
.npmrc		.npmrc
.nvmrc		.nvmrc
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
Dockerfile		Dockerfile
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
docker-compose.yml		docker-compose.yml
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
tsconfig.json		tsconfig.json
tsdown.config.ts		tsdown.config.ts
vitest.config.ts		vitest.config.ts

Folders and files

Latest commit

History

Repository files navigation

What is Nomos?

Get Running in 2 Minutes

npm (GitHub Packages)

Docker Compose (includes database)

Docker (standalone)

From source

Prerequisites

What You Get

Memory That Persists Across Sessions and Channels

Adaptive Memory & User Model

Self-Improvement — An Agent That Evolves Itself

Connect From Anywhere

Web-Based Management Dashboard

Extend Without Forking

Deploy to 6 Platforms in Minutes

Smart Model Routing — Cut Costs Without Cutting Quality

Multiple API Providers

Secrets Encrypted at Rest

Multi-Agent Teams — Parallelize Complex Work

Image & Video Generation — Built In, Not Bolted On

Beyond the Basics

Daemon Mode

How the Daemon Works

Channel Integrations

Slack

Discord

Telegram

WhatsApp

Data Integrations (via MCP)

API Providers

Skills System

Creating a custom skill

Development

Required

Provider (set one)

Model and behavior

Channel integrations

gRPC

WebSocket (Legacy)

Contributing

License

About

Topics

Resources

License

Code of conduct

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases 1

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages