Autonomous agents with a nervous system, immune system, and collective memory. Self-healing. Self-growing. Dream cycles that learn overnight. The Watcher predicts what you need before you ask.
Not a chatbot. An organism.
askalf.org · Demo · Discord · @ask_alf · Changelog
curl -fsSL https://get.askalf.org | bashThat's it. Fleet running at localhost:3001 in 60 seconds.
Your team runs 24/7. This is a real overnight cycle:
10:49 PM Watchdog catches a regression in API response times. Creates ticket.
10:51 PM Builder claims ticket, traces root cause, writes the fix. $0.14
11:03 PM Builder Tests pass. Ticket resolved. 12 min
11:14 PM Security scans 847 dependencies. Finds 2 CVEs. Patches both. $0.08
01:15 AM Builder detects API contract break from the fix. Updates it. $0.12
02:04 AM Watchdog flags Redis at 91%. Consults memory. Applies fix. → 52%
02:30 AM Fleet Chief Dream Cycle starts. Replays 48 executions. 6 patterns.
03:12 AM Fleet Chief Creates antibody for Redis pattern. Writes memory.
04:30 AM Fleet Chief Rewrites Cost Optimizer prompt (underperforming).
06:47 AM Watchdog All clear. 6 agents, 18 executions, 8 hours. $0.43
06:48 AM ── Next cycle begins. Alf briefs you when you check in.
Six biological systems. One living fleet.
Agents communicate directly — request, inform, consult, signal, handoff. Confidence and urgency signals flow across the fleet in real time. Fleet Chief auto-intervenes on critical signals.
When something breaks, agents form response teams — like white blood cells converging on an infection. After fixing, antibodies prevent the same issue from ever happening again.
Every execution grows a shared knowledge graph. 600+ nodes, 800+ edges. Concepts, patterns, decisions all linked. Agents query the graph for context before acting.
2–6 AM: the fleet replays its day, extracts patterns, writes procedural memories, consolidates duplicates, generates predictions. Issues caught before they happen.
Learns your daily patterns — when you check analytics, when you do releases, when you review costs. Pre-runs tasks 30 minutes before you ask.
Every agent scored on completion rate, cost efficiency, consistency, resolution quality. Fleet Chief promotes top performers, retrains underperformers. Natural selection.
Self-growing. Self-managing. Self-healing.
| Agent | Role | Model | Schedule | Rep |
|---|---|---|---|---|
| Fleet Chief | Meta-Agent | Sonnet | 6hr | 0.92 |
| Builder | Worker | Sonnet | Tickets | 0.83 |
| Watchdog | Monitor | Haiku | 30min | 0.81 |
| Security | Auditor | Sonnet | 6hr | 0.63 |
| Cost Optimizer | Analyst | Haiku | 12hr | 0.84 |
| Backup Agent | Ops | Haiku | Daily | 0.90 |
| + Discord, GitHub, Social, Analytics, KB Writer, Landing Page, Release Manager |
Fleet Chief creates new agents when gaps are found. No hardcoded team — infinite growth.
Cost scales with fleet size and usage. Monitors on Haiku. Workers on Sonnet. Cost Optimizer continuously adjusts model routing and schedules.
Tell Alf what you need in plain English. Alf figures out who to assign, what tools to use, and how much it'll cost.
| ⚙ Software Dev | ☁ DevOps | ☆ Marketing | ☎ Support | ⌂ E-Commerce |
| ⌘ Research | ♡ Personal | ⚒ Agency | ⚖ Finance | ★ Custom |
109 templates across 16 categories. Pre-built specialists ready in seconds.
9-step wizard: workspace type → AI provider → use case → meet your team → first task.
Tickets, executions, cost tracking, interventions.
npm i -g @askalf/agent
askalf-agent connect <key> --url ws://server:3005 --name prod-box --installAuto-detects OS. Installs as service (systemd / launchd / Windows). Runs on boot. Capabilities scan. See @askalf/agent.
| Templates | 109 across 16 categories |
| Tools | 70 (44 forge + 26 MCP) |
| Channels | 16 (Slack, Discord, Telegram, WhatsApp, Teams, Email, Twilio + 9 more) |
| Integrations | 47 across 13 categories |
| AI Models | Unlimited via Ollama (Llama, Mistral, Qwen, DeepSeek, Phi, Gemma) + Anthropic, OpenAI, xAI, Google |
| Agents | Self-growing fleet — Fleet Chief creates agents as needed |
| Nervous System | Agent-to-agent messaging, signal board, fleet awareness |
| Immune System | Self-healing response teams, antibodies, incident management |
| Collective Memory | Shared knowledge graph, dream cycles, pattern extraction |
| The Watcher | Predictive dispatch — pre-runs tasks before you ask |
| Webhook Triggers | GitHub push/issue/PR fire agents automatically |
| Team Collaboration | Invite members, role-based access |
| Scheduled Reports | Daily/weekly summaries to Discord or email |
| Federation | Opt-in anonymized learning between instances |
TypeScript 5.4 · Node.js 22 · React 19 · Fastify 5
PostgreSQL 17 · pgvector 0.8 · Redis 8 · Docker · Cloudflare
| AskAlf | AutoGPT | CrewAI | ChatGPT | |
|---|---|---|---|---|
| Self-hosted / data stays local | ✓ | ✓ | ✓ | ✗ |
| Autonomous 24/7 fleet | ✓ | ~ | ✗ | ✗ |
| Nervous system (agent-to-agent) | ✓ | ✗ | ✗ | ✗ |
| Immune system / self-healing | ✓ | ✗ | ✗ | ✗ |
| Dream cycles / overnight learning | ✓ | ✗ | ✗ | ✗ |
| Collective memory / knowledge graph | ✓ | ✗ | ✗ | ~ |
| Self-growing fleet | ✓ | ✗ | ✗ | ✗ |
| Agent reputation & natural selection | ✓ | ✗ | ✗ | ✗ |
| 109 ready-to-use templates | ✓ | ✗ | ~ | ✗ |
| Remote device execution | ✓ | ✗ | ✗ | ✗ |
From launch to organism. 10 days.
| Version | Name | What Changed |
|---|---|---|
| v2.9.5 | The Immune System | Self-healing response teams, antibodies, incident management |
| v2.9.0 | The Nervous System | Agent-to-agent communication, collective memory, signals |
| v2.8.0 | The Self-Evolving Release | Dream cycles, The Watcher, reputation, federation, webhook triggers |
| v2.7.0 | Autonomous Fleet | Fleet Chief, Builder, distributed execution |
| v2.5.0 | Remote Execution | Device bridge, smart routing, agent CLI |
| v2.2.0 | Security & Memory | Vector search, Docker hardening, 70 tools |
| v2.1.0 | Launch | 109 templates, 16 channels, persistent memory |
See full CHANGELOG.
Reports · Webhooks · Team · Integrations · Channels · Devices · Budget · Infrastructure
askalf.org · Demo · Discord · npm · @ask_alf
MIT — askalf.org