Daemora

Self-hosted AI agent platform - autonomous, multi-channel, multi-model.

Deploy once on your machine. Connect your channels. Message the bot - it writes code, does research, sends emails, runs cron jobs, and reports back. You own everything.

52 built-in tools. 20 channels. 14 security layers. 7 AI providers. Multi-agent teams. Production-grade scheduling. Continuous brain with three-layer memory. Crew system with Media Studio. MCP integration. Provider failover. Smart loop detection. All self-hosted - nothing leaves your infrastructure except the tokens you send to model APIs.

Daemora fixes a bug in a GitHub repo, opens a PR, runs the test suite locally, and pings you every minute with a health check — autonomously.

Ask Daemora to research catnip. It searches the web, synthesises a report, saves it to your machine, and sends you the file directly on Telegram.

"Find the top 10 skipping ropes on Amazon, analyse them, create a Google Doc with the results, and email it to someone." One message. Fully autonomous.

npm install -g daemora
daemora setup      # interactive wizard - models, channels, tools, cleanup, vault, MCP
daemora start      # start the agent

Then message your bot. That's it.

npm install -g daemora
daemora setup
daemora start

git clone https://github.com/CodeAndCanvasLabs/Daemora.git
cd daemora-agent
pnpm install
cp .env.example .env
# Add your API keys to .env
daemora setup
daemora start

daemora daemon install    # Register as a system service (launchctl / systemd / Task Scheduler)
daemora daemon start      # Start in background
daemora daemon status     # Check status
daemora daemon logs       # View logs
daemora daemon stop       # Stop

Copy .env.example to .env and fill in what you need.

At least one provider is required:

OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_AI_API_KEY=...
XAI_API_KEY=...
DEEPSEEK_API_KEY=...
MISTRAL_API_KEY=...

# Default model (used when no model is specified)
DEFAULT_MODEL=openai:gpt-4.1-mini

7 providers, 59+ models - including:

Use format provider:model-id (e.g. openai:gpt-5.2, anthropic:claude-sonnet-4-6). Supports dynamic passthrough for any model ID your provider accepts.

Route different task types to the best model automatically:

CODE_MODEL=anthropic:claude-sonnet-4-6
RESEARCH_MODEL=google:gemini-2.5-flash
WRITER_MODEL=openai:gpt-4.1
ANALYST_MODEL=openai:gpt-4.1
SUB_AGENT_MODEL=openai:gpt-4.1-mini    # default model for all sub-agents

When a sub-agent is spawned with profile: "coder", it automatically uses CODE_MODEL. Sub-agents without a profile route fall back to SUB_AGENT_MODEL, then the parent's model, then DEFAULT_MODEL.

Enable only what you need. Each channel supports {CHANNEL}_ALLOWLIST and {CHANNEL}_MODEL overrides.

daemora channels             # List all channels + setup status
daemora channels add         # Configure a new channel interactively
daemora channels add discord # Configure a specific channel directly

MAX_COST_PER_TASK=0.50     # Max $ per task (agent stops mid-task if exceeded)
MAX_DAILY_COST=10.00       # Max $ per day across all tasks

PERMISSION_TIER=standard           # minimal | standard | full
ALLOWED_PATHS=/home/user/work      # Sandbox: restrict file access to these directories
BLOCKED_PATHS=/home/user/.secrets  # Always block these, even inside allowed paths
RESTRICT_COMMANDS=true             # Block shell commands referencing paths outside sandbox

MCP (Model Context Protocol) lets Daemora control external tools. Each connected server gets a specialist sub-agent with focused context.

# Add a server (interactive)
daemora mcp add

# Add a server (command line)
daemora mcp add github npx -y @modelcontextprotocol/server-github
daemora mcp add notion npx -y @notionhq/notion-mcp-server
daemora mcp add myserver https://api.example.com/mcp          # HTTP
daemora mcp add myserver https://api.example.com/sse --sse    # SSE

# Manage servers
daemora mcp list              # Show all configured servers
daemora mcp enable github     # Enable a server
daemora mcp disable github    # Disable without removing
daemora mcp reload github     # Reconnect after config changes
daemora mcp remove github     # Remove permanently

Popular MCP servers:

Crew members are self-contained specialist sub-agents. Each has its own tools, profile, skills, and persistent session. The main agent delegates via useCrew(crewId, task).

daemora crew install daemora-crew-weather
daemora crew list
daemora crew remove weather

See crew/README.md for the full guide. Three files:

crew/my-crew/
├── plugin.json       # manifest (id, name, description, profile, skills)
├── index.js          # register tools via api.registerTool()
└── tools/
    └── myTool.js     # tool implementation

daemora crew list         # Show all crew members + status
daemora crew install <pkg> # Install from npm
daemora crew remove <id>   # Remove a crew member
daemora crew reload        # Hot-reload all crew members

52 tools the agent uses autonomously:

Teams enable coordinated multi-agent collaboration with shared task lists, dependency tracking, and inter-agent messaging. Use teams when agents need to coordinate - use parallelCrew when they don't.

createTeam("feature-sprint")
  → addTeammate({ profile: "coder", instructions: "..." })
  → addTeammate({ profile: "researcher" })
  → addTeammate({ profile: "writer" })
  → addTask({ title: "Research API options", ... })
  → addTask({ title: "Implement chosen API", blockedBy: ["task-1"] })
  → addTask({ title: "Write docs", blockedBy: ["task-2"] })
  → spawnAll()
  → teammates claim tasks, work, complete, message each other
  → disbandTeam()

• Max 5 teams active
• Max 10 teammates per team
• Max 7 concurrent sub-agents
• Max nesting depth: 3

Skills inject behaviour instructions when a task matches certain keywords. Create a .md file in skills/ with a YAML frontmatter:

---
name: deploy
description: Handle deployment tasks for web apps and APIs
triggers: deploy, release, ship, production, go live
---

# Deployment Checklist

Always follow this order when deploying:

1. Run the full test suite - never deploy broken code
2. Check for .env differences between dev and prod
3. Build the production bundle
4. Use zero-downtime deployment if possible (blue/green, rolling)
5. Verify the deployment is healthy before reporting done
6. Notify the user with the live URL

60 built-in skills cover: coding, research, email, weather, Spotify, Obsidian, Apple Notes, Apple Reminders, Things, Trello, Tmux, PDF, image generation, video editing (Remotion), video frames, music generation, health checks, GIF search, webcam capture, documents (PDF/DOCX/XLSX/PPTX), data analysis, DevOps, API development, browser automation, meeting attendance, planning, orchestration, GitHub, Discord ops, Slack ops, Google Workspace, macOS automation, and more.

# Run a full security audit
daemora doctor

SQLite database (data/daemora.db) stores configuration, sessions, tasks, cron jobs, vault secrets, channel identities, and three-layer memory (semantic/episodic/procedural). File-based storage is used for audit logs and cost tracking.

data/
├── daemora.db      SQLite database (config, sessions, tasks, vault, cron, memory, learning_log)
├── audit/          Append-only JSONL audit logs (secrets stripped)
├── costs/          Per-day cost tracking logs
└── skill-embeddings.json

Configurable retention prevents unbounded growth. Set via CLEANUP_AFTER_DAYS env var, CLI, or setup wizard.

daemora cleanup stats            # Show storage usage
daemora cleanup set 30           # Auto-delete files older than 30 days
daemora cleanup set 0            # Never auto-delete
daemora cleanup                  # Run cleanup now

Auto-cleanup runs on startup. Cleans: tasks, audit logs, cost logs, and stale sub-agent sessions. Main user sessions are never auto-deleted.

daemora start                    Start the agent server
daemora setup                    Interactive setup wizard
daemora doctor                   Security audit - scored report

daemora mcp list                 List all MCP servers
daemora mcp add                  Add an MCP server (interactive)
daemora mcp add <name> <cmd>     Add an MCP server (non-interactive)
daemora mcp remove <name>        Remove an MCP server
daemora mcp enable <name>        Enable a disabled server
daemora mcp disable <name>       Disable without removing
daemora mcp reload <name>        Reconnect a server

daemora daemon install           Install as a system daemon
daemora daemon start             Start the daemon
daemora daemon stop              Stop the daemon
daemora daemon status            Check daemon status
daemora daemon logs              View daemon logs

daemora vault set <key>          Store an encrypted secret
daemora vault get <key>          Retrieve a secret
daemora vault list               List all secret keys
daemora vault unlock             Unlock the vault

daemora sandbox show             Show current sandbox rules
daemora sandbox add <path>       Allow a directory (activates scoped mode)
daemora sandbox remove <path>    Remove from allowed list
daemora sandbox block <path>     Always block a path
daemora sandbox restrict         Enable command restriction
daemora sandbox clear            Back to unrestricted mode

daemora channels                 List all channels + setup status
daemora channels add             Configure a new channel interactively
daemora channels add <name>      Configure a specific channel directly

daemora cleanup                  Run data cleanup now (uses configured retention)
daemora cleanup stats            Show storage usage (tasks, sessions, audit, costs)
daemora cleanup set <days>       Set retention period (0 = never delete)

daemora help                     Show full help

The agent exposes a REST API on http://localhost:8081.

# System health
curl http://localhost:8081/health

# List recent tasks
curl http://localhost:8081/tasks

# Get task status
curl http://localhost:8081/tasks/{taskId}

# Today's API costs
curl http://localhost:8081/costs/today

# List MCP servers
curl http://localhost:8081/mcp

POST /chat and POST /tasks (unauthenticated task submission) are disabled by default - use a channel (Telegram, Slack, etc.) instead.

Daemora runs entirely on your own machine. Nothing is sent to any third party beyond the AI model APIs you configure.

Requirements:

• Node.js 20+
• 512 MB RAM minimum
• macOS, Linux, or Windows WSL

Production setup:

npm install -g daemora
daemora setup
daemora daemon install
daemora daemon start
daemora doctor       # verify security configuration

Use nginx or Caddy as a reverse proxy for HTTPS if exposing the API port.

pnpm test                  # Run all tests
pnpm test:watch            # Interactive watch mode
pnpm test:coverage         # Coverage report
pnpm test:unit             # Unit tests only
pnpm test:integration      # Integration tests only

97 tests covering: Task lifecycle, CostTracker (daily budgets), SecretScanner (pattern + blind env-var redaction), FilesystemGuard (blocked patterns, path scoping), ModelRouter (task-type routing, profile resolution), and integration tests.

git clone https://github.com/CodeAndCanvasLabs/Daemora.git
cd daemora-agent
pnpm install
cp .env.example .env
# Add your API keys to .env
daemora setup
pnpm test          # Make sure everything passes
daemora start

Contributions are welcome. Please open an issue before submitting large PRs.

AGPL-3.0 - Daemora is open source. If you modify Daemora and distribute it, or run it as a network service, you must open-source your changes under AGPL-3.0.

See LICENSE for the full text.

• Website: https://daemora.com
• npm: https://npmjs.com/package/daemora
• GitHub: https://github.com/CodeAndCanvasLabs/Daemora
• Issues: https://github.com/CodeAndCanvasLabs/Daemora/issues