Tell Alf what you need. Alf builds the team.
One AI that creates specialized workers for any task — marketing, support, e-commerce, research, finance, personal productivity, whatever you need. Alf figures out who to hire, puts them to work, and keeps everything running 24/7.
Other tools give you a chatbot. AskAlf gives you a workforce.
askalf.org · Discord · Discussions · @ask_alf
curl -fsSL https://get.askalf.org | bashThat's it. Checks prerequisites, generates secrets, pulls images, and starts your team at http://localhost:3001.
Runs on Linux, macOS, and Windows WSL2 — anything with Docker.
Manual install
git clone https://github.com/askalf/askalf.git
cd askalf && ./setup.sh
docker compose -f docker-compose.selfhosted.yml up -dYour team runs 24/7. Here's a real cycle:
- 10:49 PM — A monitor catches a regression. Creates a ticket.
- 10:51 PM — A builder claims it, traces the root cause, writes the fix. 12 minutes later.
- 11:03 PM — Tests re-run. All pass. Ticket resolved.
- 11:14 PM — Security scans 847 dependencies. Finds 2 CVEs. Patches both.
- 01:15 AM — Another worker detects an API contract break from the fix. Updates it.
- 02:04 AM — Monitor flags Redis at 91%. Consults memory for past fixes. Applies it. Redis drops to 52%.
- 03:12 AM — A Compliance Auditor is spawned on demand to check regulatory drift. Cost: $0.04.
- 04:30 AM — Writer generates a status report.
- 06:47 AM — Cycle complete. 6 workers, 18 executions, 8 hours. Total cost: $0.43.
- Next cycle begins automatically. Alf briefs you whenever you check in.
Tell Alf what you need. Alf creates the right specialist — for any industry, any task.
- You describe what you need in plain English
- Alf figures out what kind of specialist is required
- If one exists on your team, Alf assigns them. If not, Alf creates one.
- The specialist executes with the right tools, system prompt, and domain knowledge
- Results are stored in memory — next time is faster
Competitor Researcher · SEO Analyst · Customer Support Agent · Invoice Monitor · Content Writer · Security Scanner · Data Analyst · Compliance Auditor · Social Media Monitor · Review Responder · Report Generator · Meal Planner · Travel Researcher · Habit Tracker · Budget Coach · Pet Care Scheduler · and any specialist you need
109 built-in templates across 16 categories. Workers are first-class — they create tickets, store memories, learn from experience, and coordinate with each other.
Browse, share, and install worker templates from the community:
- Submit your custom templates to the public library
- Install community-created skills with one click
- Rate and review templates to surface the best ones
- Import/Export skill bundles as JSON — share entire packs with your team
- Alf-curated featured skills reviewed and approved by the platform
Full Claude Code and OpenAI Codex terminal sessions embedded directly in the dashboard. Not wrappers — real PTY sessions via xterm.js with your full project available.
- Claude Code — with MCP tools, your knowledge graph, and full platform context
- OpenAI Codex — with dynamic instructions injected from the platform
- Switch between them with a tab. Same toolbar. Same workspace.
Every task, every outcome, every interaction is stored in a cognitive memory system that grows with every execution.
- Semantic Memory — Facts, concepts, system architecture (pgvector embeddings)
- Episodic Memory — What happened, what worked, what failed
- Procedural Memory — Successful patterns extracted into reusable templates
- Knowledge Graph — 1,500+ nodes with automatic cross-agent connection discovery
- Metabolic Consolidation — Continuous cycles that strengthen useful memories and decay noise
The 10th task is faster than the 1st. The 100th research report knows exactly where to look.
Your workers don't just run in the cloud — they control real machines.
| Category | Devices |
|---|---|
| Compute | CLI Agent, Docker Host, SSH Remote, Kubernetes |
| Desktop & Mobile | Browser Bridge, Desktop Control, VS Code, Android, iOS |
| IoT & Edge | Raspberry Pi, Arduino/ESP32, Home Assistant |
Slack · Discord · Telegram · WhatsApp · Teams · OpenClaw · REST API · Webhooks · Zapier · n8n · Make · Email · Twilio SMS · SendGrid · Twilio Voice · Zoom
| Category | Integrations |
|---|---|
| Source Control | GitHub · GitLab · Bitbucket |
| CRM | Salesforce · HubSpot · Pipedrive |
| E-Commerce | Shopify · Stripe · WooCommerce · Square |
| Marketing | Mailchimp · Google Ads · Meta Ads · SendGrid |
| Social | Twitter/X · Instagram · LinkedIn · Buffer |
| Productivity | Google Workspace · Microsoft 365 · Airtable · Google Sheets |
| Cloud | AWS · GCP · Azure · DigitalOcean |
| CI/CD | Vercel · Netlify · Railway · Fly.io |
| PM | Jira · Linear · Notion · Asana |
| Monitoring | Datadog · Sentry · PagerDuty · Grafana |
| Analytics | Google Analytics · Mixpanel · Plausible |
| Finance | QuickBooks · Xero · Gusto · Wise |
| Storage | Cloudflare · S3 · Supabase |
| Provider | Auth |
|---|---|
| Anthropic (Claude) | OAuth PKCE + API key |
| OpenAI | API key + Codex device auth |
| Ollama | Local models, no API key (Llama, Mistral, Phi, Qwen, CodeLlama, Gemma, etc.) |
| xAI (Grok) | API key |
| DeepSeek | API key |
This is not a framework. Not a library. Not a toolkit you assemble yourself.
| AskAlf | OpenClaw | Frameworks (CrewAI, AutoGen, LangGraph) | |
|---|---|---|---|
| Multi-specialist teams | Unlimited, built on demand | Single agent | You build it |
| Dashboard | Mission control + Alf chat | CLI/chat only | You build it |
| Memory | 10-layer cognitive brain with pgvector | 24h context window | You build it |
| Deployment | curl | bash, 60 seconds |
npm install -g |
You build it |
| Orchestration | Autonomous ticket dispatch | Reactive only | You build it |
| Security | AES-256, sandboxed, VPN, audited | 512 vulnerabilities found | You build it |
| Cost tracking | Per-agent budgets + audit log | Token estimates | You build it |
| Channels | 16 built in (inc. OpenClaw) | 19 chat platforms | You build it |
| Marketplace | 109 templates + community library | ClawHub skills | You build it |
| AI terminals | Embedded Claude CLI + Codex | Not available | Not available |
| VPN tunneling | Built-in Gluetun + Proton VPN | Not available | Not available |
| Auto-recovery | Autoheal container self-healing | Not available | Not available |
┌─────────────────────────────────────────────────────┐
│ Dashboard │
│ Ask Alf · Team · Workspace · Ops · Memory · Marketplace │
│ Claude Code · Codex · Settings │
├───────────────┬───────────────┬──────────────────────┤
│ Forge │ MCP Tools │ SearXNG │
│ API Server │ Agent Tools │ Web Search │
│ Runtime │ Memory Ops │ │
│ Scheduler │ Ticket Ops │ │
├───────────────┴───────────────┴──────────────────────┤
│ PostgreSQL + pgvector │ Redis │
└─────────────────────────────────────────────────────┘
TypeScript 5.4 · Node.js 22 · React 19 · Fastify 5 · PostgreSQL 17 · pgvector 0.8 · Redis 8 · Docker Compose · xterm.js · node-pty · WebSocket · MCP Protocol · PKCE OAuth · SearXNG · Gluetun VPN · Autoheal · Ollama (local models)
Route all outbound agent traffic through an encrypted VPN tunnel via Gluetun. Your agents' API calls, web searches, and external requests stay encrypted and anonymous.
# Add to your .env file
PROTON_WIREGUARD_PRIVATE_KEY=your-key-from-protonvpn-dashboard
PROTON_SERVER_COUNTRIES=United States
# VPN is included by default — all search and outbound traffic routes through Gluetun
docker compose -f docker-compose.selfhosted.yml up -dSupports 30+ providers — ProtonVPN, Mullvad, NordVPN, Surfshark, and more. Change the country to route through any location your provider supports.
Two ecosystems built into the dashboard:
MCP Tool Marketplace — 44 forge tools + 26 MCP tools = 70 tools (tickets, findings, Docker, deploy, security scan, code analysis, knowledge graph, twitter_ops, discord_ops, fleet health, git ops, and more). Install community-published tools with one click.
Community Skills Library — 109 built-in templates across 16 categories (Personal, Marketing, Support, E-Commerce, Content, Finance, Legal, HR, Operations, Research, and more). Submit your own skills, browse community submissions, import/export bundles, and install Alf-curated featured templates.
Already running OpenClaw? Connect it as a channel frontend to AskAlf. Messages from OpenClaw-connected platforms (WhatsApp, Telegram, Discord, etc.) route through your AskAlf team with full memory, orchestration, and coordination.
# Add to your .env
OPENCLAW_GATEWAY_URL=ws://127.0.0.1:18789
OPENCLAW_GATEWAY_TOKEN=your-gateway-tokenOpenClaw handles the chat. AskAlf handles the thinking.
Compare features: AskAlf Security Policy
One-command migration converts your OpenClaw agents, skills, memory, and config to AskAlf format:
./scripts/migrate-from-openclaw.sh ~/.openclaw
# Review: ./openclaw-migration/
# Import: ./openclaw-migration/import-to-askalf.shConverts: agents (AGENTS.md + SOUL.md → system prompts), skills (SKILL.md → templates), memory (MEMORY.md → semantic seeds), channels, providers, and heartbeat config.
Automatic container recovery — included by default. If any container fails its health check, Autoheal restarts it automatically.
How much does it cost to run?
Free on your own hardware. AI provider costs depend on usage — typical 24/7 operation with 8 workers costs $0.30–$2.00/day with Claude. Use Ollama for fully free local operation.
Can I run it without cloud AI providers?
Yes. Configure Ollama as your provider and run Llama, Mistral, or any local model. No API keys needed.
How is this different from CrewAI / AutoGen / LangGraph?
Those are frameworks — you write code to wire agents together. AskAlf is a complete platform: dashboard, scheduler, memory, marketplace, and runtime. You describe what you need in plain English; Alf builds the team.
Is my data safe?
Everything runs on your hardware. AES-256-GCM encryption at rest, all credentials encrypted, VPN tunneling for outbound traffic, HSTS + CSP on all endpoints. Your data never leaves your infrastructure.
What models does it support?
Anthropic Claude (Opus, Sonnet, Haiku), OpenAI (GPT-4o, o3, Codex), xAI Grok, DeepSeek, and any Ollama-compatible model (Llama, Mistral, Phi, Qwen, CodeLlama, Gemma).
More questions? Ask in GitHub Discussions.
- Docker and Docker Compose
- 4GB+ RAM (8–12GB recommended for full fleet)
- At least one AI provider: Anthropic (recommended), OpenAI, xAI, DeepSeek, or Ollama for fully local/free operation
- Free to run on your own hardware — or ~$5/month on a VPS
MIT — see LICENSE
Built by askalf · askalf.org · support@askalf.org