DAEMORA
Does the work for you
not with you holding its hand
Daemora connects to any AI model, any tool, any channel. It plans, codes, tests, researches, emails, and ships - without asking permission at every step.
npm install -g daemoraHow Daemora Works
Three simple steps to autonomous AI execution
Install
Install the package globally with npm. Run the setup wizard. Add your API keys and connect your tools.
Connect Your Tools
Connect AI models (OpenAI, Anthropic, Google, Ollama), messaging channels (Telegram, WhatsApp, Discord, Slack, Email, LINE, Signal, Teams, Google Chat), and MCP servers (Notion, GitHub, Linear, and thousands more).
Give It Work
Send a task via Telegram, WhatsApp, Slack, Discord, Signal, Teams, Google Chat, or any other connected channel. Daemora plans it, executes it with tools, verifies the result, and reports back.
Total setup time: ~2 minutes
Everything in One Agent
No assembly required. Daemora comes loaded with everything you need.
Autonomous Coding
Writes code, runs builds, starts dev servers, takes screenshots to verify the UI, writes test cases, fixes failures. Ships working software without being hand-held.
Web Research
Searches the web, reads full pages, analyzes images and charts, cross-references sources. Delivers findings as a structured report, not a list of links.
MCP Tool Control
Connect any MCP server - Notion, GitHub, Linear, Slack, Shopify, PostgreSQL, and more. Daemora calls them autonomously to complete tasks.
22 Live Channels
Receive tasks over 22 channels - Telegram, WhatsApp, Discord, Slack, Email, Teams, iMessage, Matrix, Mattermost, Twitch, Nostr, IRC, Nextcloud, Zalo, Feishu, and more. Per-channel model override, allowlists, and status reactions built in.
Multi-Agent Orchestration
Spawn parallel sub-agents for different parts of a task. A researcher, coder, and writer working at the same time - with cascade kill and mid-run steering.
Scheduling & Automation
Schedule recurring tasks with cron expressions. Add, update, enable, or disable schedules at runtime. Daily reports, nightly builds - no triggers needed.
Voice & Media Pipeline
Send voice notes - Daemora transcribes them via Whisper and acts on them. Photos, videos, and docs are processed and attached. Can send files and audio back too.
Text-to-Speech
Respond with audio via OpenAI tts-1-hd (6 voices) or ElevenLabs. Daemora can read back reports, summaries, and results - not just text.
Semantic Memory
Long-term MEMORY.md with vector search via OpenAI embeddings and cosine similarity. Per-tenant isolated memory - no cross-user leaks. Remembers context across sessions.
Multi-Tenant Architecture
Run one instance for your whole team. Each user gets isolated filesystem, memory, API keys, model tier, cost limits, tool allowlists, and MCP access - fully sandboxed.
10-Layer Security
Permission tiers, filesystem sandbox, secret vault (AES-256-GCM), channel allowlists, audit log, supervisor agent, input sanitization, A2A auth, tenant isolation, and security audit CLI.
Security Audit CLI
Run `daemora doctor` for an 8-check scored security audit. Checks vault, allowlists, sandbox, tenant isolation, cost limits, and A2A config - with actionable fix instructions.
21 Built-in Skills
Weather, image generation, PDF, Spotify, Obsidian, Apple Notes, Reminders, Things, Trello, tmux, video frames, GIF search, healthcheck, and more. Routed via semantic cosine similarity.
Hook System
PreToolUse and PostToolUse hooks let you intercept any tool call. Configure in hooks.json to log, modify, or block tool executions without touching agent code.
A2A Protocol
Agent-to-agent task delegation with bearer auth, URL allowlist, rate limiting, and a separate cost budget. Expose your agent to other agents securely via /.well-known/agent.json.
Auto Task Recovery
Tasks in-flight when the agent crashes are automatically re-queued on restart. No manual re-input needed - completed tasks are routed back to their originating channel.
One Instance. Every User Isolated.
Run Daemora for your entire team. Each user gets their own sandboxed environment - memory, filesystem, API keys, model, cost budget, and tool access - all isolated at the runtime level.
Filesystem Isolation
Each tenant is scoped to their own directories. Hardcoded blocks on sensitive paths (.ssh, .env, .aws). CLI-managed via `daemora sandbox`.
Memory Isolation
Separate MEMORY.md, daily logs, and vector embeddings per user. Memory is never shared or leaked across tenants - even in concurrent requests.
Per-Tenant API Keys
AES-256-GCM encrypted keys stored per tenant. Decrypted at task start and passed through the call stack - never written to process.env.
Model & Routing
Each tenant gets their own model tier. Free users get a cheap model, pro users get Claude Opus. Task-type routing: CODE_MODEL, RESEARCH_MODEL, WRITER_MODEL.
Cost Limits
Per-task and daily cost caps enforced per user. Usage tracked independently in the cost log. Free tier and pro tier get different budgets.
Tool & MCP Allowlists
Restrict which tools (including MCP tools) and which MCP servers each tenant can access. Null = all. Or enumerate specific allowed names.
Tenant Suspension
Instantly suspend any user with one command. All their tasks are rejected at dequeue time. Survives agent restarts. Includes a reason field.
AsyncLocalStorage Context
Every task runs inside tenantContext.run(). All tools read from the running context - not a global - so concurrent tasks from different users never bleed.
Manage Tenants via CLI
Full CRUD from the command line. No database required - stored as flat JSON.
Control Any Tool via MCP
MCP (Model Context Protocol) is an open standard that lets Daemora plug into thousands of external tools. Add a server in seconds - Daemora handles the rest.
+ Thousands more MCP-compatible servers
Any MCP-compatible server works. If it speaks the MCP protocol, Daemora can use it.
daemora mcp addRuns an interactive prompt to add any server
Works Where You Work
19 live channels - Telegram, WhatsApp, Discord, Slack, Email, Teams, Google Chat, LINE, Signal, Matrix, Mattermost, Twitch, IRC, iMessage, Feishu, Nextcloud Talk, BlueBubbles, Nostr, and more. Choose what fits your workflow.
Telegram
Create a bot, add your token, and message Daemora directly from Telegram. Supports long messages, split responses, approval flows, voice note transcription, and file sharing.
Connect via Twilio. Send tasks over WhatsApp - your most-used app becomes a command interface. Supports voice note transcription and file attachments.
Discord
Connect to Discord servers. Use slash commands or DMs. Supports voice message transcription, status reactions (⏳ ✅ ❌), and per-channel model override.
Slack
Mention @daemora in any channel or DM it directly. Reacts with status emoji, supports per-channel model selection, and allowlists for workspace security.
Connect any IMAP/SMTP account. Email a task, get a reply when it's done. Works with Gmail, Outlook, Fastmail - any standard email provider.
Microsoft Teams
Connect via Bot Framework v4 CloudAdapter. DM or mention in channels. Per-channel model override and allowlist support. File attachments supported.
Google Chat
Connect via OIDC service account authentication. DM or mention in Spaces. Per-channel model override and allowlist support.
LINE
Integrate via LINE Messaging API with HMAC webhook verification. Full allowlist support and per-channel model override for teams across Asia using LINE.
Signal
Connect via signal-cli REST. End-to-end encrypted, allowlist-gated. The most private channel - ideal for security-conscious users and sensitive workflows.
Matrix
Connect via access token to any Matrix homeserver (Element, Synapse). Allowlist-gated, per-channel model override. Federated and self-hostable.
Mattermost
Connect via bot token over WebSocket. Mention in channels or DM directly. Allowlist support and per-channel model override for self-hosted teams.
Twitch
Connect via OAuth IRC. Receive tasks from Twitch chat, respond in channel. Allowlist-gated so only trusted users can trigger the agent.
IRC
Connect to any IRC server via NickServ auth. Classic protocol, zero dependencies. Allowlist-gated, per-channel model override.
iMessage
Connect via AppleScript on macOS. Send tasks from iMessage - your Apple devices become a command interface. Allowlist-gated by phone number or Apple ID.
Feishu / Lark
Connect via Feishu/Lark App ID and Secret. DM or mention in group chats. Popular with Asia-Pacific teams - full allowlist and model override support.
Nextcloud Talk
Connect via app password to your self-hosted Nextcloud instance. Fully private - no data leaves your server. Allowlist and model override supported.
BlueBubbles
Connect via BlueBubbles server password. iMessage on non-Apple hardware - send tasks from any device via your Mac relay server.
Nostr
Connect via private key using NIP-04 encrypted DMs. Fully decentralized, censorship-resistant. The most sovereign channel for users who own their infrastructure.
Get Started in 3 Commands
Getting Daemora running on your machine takes less than 2 minutes
Install Daemora
npm install -g daemoraRequires Node.js 20 or newer.
Don't have Node.js? Download it from nodejs.org →
Run Interactive Setup
daemora setupThe interactive wizard will configure:
Start the Agent
daemora start🚀Daemora is running. Message it on any connected channel.
Run as a Background Service (optional)
Keep Daemora always running in the background, even after reboot:
Install and start daemon:
daemora daemon install
daemora daemon startCheck status:
daemora daemon statusStop daemon:
daemora daemon stopThe Right Agent for Every Task
Daemora automatically picks the right tool set based on the task. For complex work, it spawns focused sub-agents in parallel.
Researcher
Searches the web, reads documents, analyzes images, queries external sources via MCP. Delivers structured findings, not raw links.
Coder
Full ownership of code tasks. Writes, runs, tests, screenshots the UI, fixes bugs, and ships. Never hands work back incomplete.
Writer
Produces polished documents, reports, and content. Reads existing files for context, researches as needed, delivers the final output.
Analyst
Runs data processing scripts, analyzes charts and visuals, queries databases, produces insights.
Works with Every Major AI Provider
Your choice of model, your choice of provider. Switch per-task or chain models together for best results.
OpenAI
GPT-4.1, GPT-4.1 Mini, o3
Claude
Claude Opus 4, Claude Sonnet 4
Google Gemini
Gemini 2.5 Pro, Gemini 2.0 Flash
Ollama
Llama, DeepSeek, Phi (local)
Groq
Llama 3.3, Gemma 2, Mixtral
Mistral
Mistral Large, Mistral Small, Codestral
Smart Model Switching
Switch models per-task. Use a powerful model for complex reasoning, a fast cheap model for simple tasks. Daemora has automatic fallback chains - if your preferred model is unavailable, it falls back gracefully.
Developers Love Daemora
Real feedback from engineers who replaced their task list with a single AI agent.
"I gave Daemora my entire backlog of Jira tickets at 11pm. Woke up to a Telegram message saying 6 tasks were done, PRs opened, and tests passing. This thing is insane."
"Set up the Slack integration in 2 minutes. Now I just DM @daemora 'summarize all PRs from last week' and it does it with full context. No more context switching."
"The multi-agent thing is real. Asked it to research competitors AND rewrite my docs AND fix a bug - it literally spawned 3 sub-agents and ran them in parallel. Finished faster than one task normally takes me."
"Connected my PostgreSQL via MCP. Now I just tell Daemora what analysis I need over Telegram and it writes the query, runs it, and sends back a formatted report. My boss thinks I'm a wizard."
"I gave Daemora my entire backlog of Jira tickets at 11pm. Woke up to a Telegram message saying 6 tasks were done, PRs opened, and tests passing. This thing is insane."
"Set up the Slack integration in 2 minutes. Now I just DM @daemora 'summarize all PRs from last week' and it does it with full context. No more context switching."
"The multi-agent thing is real. Asked it to research competitors AND rewrite my docs AND fix a bug - it literally spawned 3 sub-agents and ran them in parallel. Finished faster than one task normally takes me."
"Connected my PostgreSQL via MCP. Now I just tell Daemora what analysis I need over Telegram and it writes the query, runs it, and sends back a formatted report. My boss thinks I'm a wizard."
"Open source AND works with local Ollama models means I can run this completely air-gapped. For my client work this was a dealbreaker requirement. Daemora is the only agent that delivers."
"I use it via WhatsApp. While commuting I sent 'write a cold email sequence for our new product'. By the time I got to the office it had written 5 emails, subject lines, and sent me a draft to review."
"The hook system is underrated. I set a PostToolUse hook that logs every file it touches to our audit trail. Now I have full visibility into what the agent did. Security team actually approved it."
"Was skeptical of the 'autonomous' claim. Then I asked it to scrape a competitor's pricing page, compare to ours, and update our Notion docs. It did all three steps without asking me anything. Believer now."
"Open source AND works with local Ollama models means I can run this completely air-gapped. For my client work this was a dealbreaker requirement. Daemora is the only agent that delivers."
"I use it via WhatsApp. While commuting I sent 'write a cold email sequence for our new product'. By the time I got to the office it had written 5 emails, subject lines, and sent me a draft to review."
"The hook system is underrated. I set a PostToolUse hook that logs every file it touches to our audit trail. Now I have full visibility into what the agent did. Security team actually approved it."
"Was skeptical of the 'autonomous' claim. Then I asked it to scrape a competitor's pricing page, compare to ours, and update our Notion docs. It did all three steps without asking me anything. Believer now."
"The cost tracking saved me. Previous agentic tools burned $40 in one runaway session. Daemora hit my $2 daily limit, stopped, and messaged me on Telegram asking if I wanted to continue. Perfect."
"AGPL open source is huge for us. We could audit every line before deploying it in our infra. The codebase is clean and well-structured. Already shipped 2 internal forks of it."
"I run it as a daemon on a $5 VPS. It monitors my GitHub issues, triages new ones, assigns labels, and pings me on Telegram for anything critical. Full-time assistant for $5/month + API costs."
"The MCP ecosystem is the real moat. Notion + Linear + GitHub all connected. I just describe what I want to happen across all three systems and Daemora coordinates it. Saved ~3 hours a day."
"The cost tracking saved me. Previous agentic tools burned $40 in one runaway session. Daemora hit my $2 daily limit, stopped, and messaged me on Telegram asking if I wanted to continue. Perfect."
"AGPL open source is huge for us. We could audit every line before deploying it in our infra. The codebase is clean and well-structured. Already shipped 2 internal forks of it."
"I run it as a daemon on a $5 VPS. It monitors my GitHub issues, triages new ones, assigns labels, and pings me on Telegram for anything critical. Full-time assistant for $5/month + API costs."
"The MCP ecosystem is the real moat. Notion + Linear + GitHub all connected. I just describe what I want to happen across all three systems and Daemora coordinates it. Saved ~3 hours a day."
Open Source · Built to Last
Daemora is licensed under AGPL-3.0 - free to use, free to inspect, free to contribute. If you modify it and run it as a service, you share your changes too. No hidden code. No black box.
Free for personal use
Install and run as many instances as you want, no cost
Community-driven
All improvements are shared back to everyone
Self-hosted
Your data stays on your machine. Nothing is sent to us.