Not a chatbot. An organism.

What do you need?

Type anything. For anyone. Alf builds the team.

Your fleet.

What makes it alive

Six biological systems. One living fleet.

ALF NERVOUS SYSTEM IMMUNE MEMORY DREAM CYCLES THE WATCHER NATURAL SELECTION
NERVOUS SYSTEM

The fleet doesn't use tickets. It feels.

Builder is stuck. Confidence: 0.3. Urgency: 0.9.

The signal propagates in 200ms. Fleet Chief receives it.
Intervention deployed. Resolution in 4 minutes.

Five message types: REQUEST, INFORM, CONSULT, SIGNAL, HANDOFF.
Real-time confidence and urgency flow across every agent.
Agents communicate directly. No queues. No polling. Synapses.
IMMUNE SYSTEM

It heals itself. Then makes sure it never happens again.

02:04 AM — Redis memory at 91%.

Watchdog detects the anomaly. Immune response triggered.
Builder and Memory join the response team.
Fix applied. Memory drops to 52%.

Antibody created: “When Redis > 85%, preemptive flush.”
Next time, the fix is automatic.
Detection, triage, mobilize, fix, verify, immunize. Six stages. Zero humans.
COLLECTIVE MEMORY

Every execution makes the next one smarter.

Builder resolves a database connection timeout.

Memory creates nodes: [Timeout][API Service][Pool Exhaustion].
Edges form: causes, fixes, relates_to.

Next time any agent approaches database issues,
it queries the graph. Context before action.
A knowledge graph that grows with every decision. Shared across every agent.
DREAM CYCLES

While you sleep, the fleet is learning.

2:00 AM — Dream Cycle begins.

REPLAY — 48 executions from the past 24 hours.
EXTRACT — 6 patterns identified.
CONSOLIDATE — duplicate memories merged.
PREDICT — 2 issues flagged before they happen.
LEARN — procedural memories written.
REPORT — fleet-wide insight stored.
Six stages. Every night. The fleet wakes up smarter than it went to sleep.
THE WATCHER

It knows what you need before you ask.

You check GitHub analytics every Monday at 9:05 AM.

Week 1: Watcher observes. Records pattern.
Week 2: Confidence rises to 75%.
Week 3: Monday, 8:35 AM — Watcher pre-runs analytics.

You log in at 9:05. The report is already waiting.
Zero latency. Zero effort.
Learns your daily patterns. Pre-runs tasks 30 minutes before you ask.
NATURAL SELECTION

The fleet evolves itself.

Every agent scored on four axes:
Completion rate · Cost efficiency · Consistency · Resolution quality

Cost Optimizer: reputation 0.35. Flagged. Three cycles in a row.
Auto-decommissioned.

Fleet Chief creates replacement with improved prompt.
New agent: reputation 0.78 after first cycle.
Top performers promoted. Underperformers replaced. Darwinian pressure on every agent.

Always running.

Pick any 8-hour window. This is what you'll find.

8 agents · 24 executions · 8 hours · $0.71 · one cycle of many

Tell it what you need.

Two ways to run. Zero excuses.
Standalone — no Docker
npx create-askalfcopied
Full platform in a single Node.js process. PGlite, in-memory cache, zero dependencies.
Docker — production stack
curl -fsSL https://get.askalf.org | bashcopied
PostgreSQL, Redis, Ollama, SearxNG. Full stack at localhost:3001 in 60 seconds.