v0.1 · 40+ modules live · chain height 75+

The operating system
for AI agents.

Every tool call signed. Every decision ledgered. Every workflow resumable. Every poor choice replayable. Cryptographic audit trail for AI ops — so when your agent screws up, you know exactly when, why, and how to undo it.

Request access → See modules
40+
Modules
30+
Agents in prod
100%
Signed events
Replayable
// What ZiggyOS is

Production runtime for autonomous agents.

Beneath every NexusFleet agent runs ZiggyOS — a deterministic, append-only, cryptographically-anchored OS. Built for agents that need to ship work, not chat.

📜

Append-only ledger

Every tool call, decision, retry, side-effect → signed event → chained block. Hash-anchored. Tamper-evident.

🔄

Saga + compensation

Multi-step external workflows with rollback paths. No half-done state. No orphaned side effects.

📸

Snapshot + rollback

Capture exact agent state. Restore to any checkpoint. Resume mid-pipeline after crash or rollback.

🧠

Memory tiers

Hot / warm / cold + episodic + semantic + procedural. Vector-indexed. Recall by similarity, not keyword.

🔑

Idempotency keys

Side effects guaranteed at-least-once, at-most-once where required. Retry storms drained automatically.

⚖️

Multi-agent voting

Council layer for consensus. N-of-M approval gates for sensitive ops (refunds, deploys, deletes).

🎬

Replay trajectories

Replay any agent run end-to-end as video / step diagram. Debug what the model actually thought.

💰

Cost tracking

Token + dollar accounting per task / goal / project / day. Hard budget caps. Cost graphs auto-generated.

🪝

Hot-reload plugins

Drop a JS file in /plugins. ZiggyOS picks it up. No restart. Marketplace + federation built-in.

"Ledger. Saga. Snapshot. Recover."
// API

Drop-in for any agent.

One HTTP endpoint. Two-line integration. Falls back silently if OSD offline — never blocks the caller.

// log any agent event to ZiggyOS const osd = require('/opt/ziggy-osd/log'); osd.emit('sm-match', 'lead.matched', { lead_id: '1778596378619-r9k7it', matches: 3, budget: { min: 800, max: 2000 }, }, { trace_id, cost_usd: 0.012, tokens_in: 1840 }); // fire-and-forget. signed. ledgered. cost-tracked. queryable.
// Inside the OS

40+ battle-tested modules.

Not a framework. An OS. Every module solves one real failure mode that took down a production agent at least once.

ledger
saga
snapshot
rollback
dlq
retry
idempotency
webhook-sign
schema-validate
memtier
vec
kgraph
autonomy
voting
council
federation
marketplace
plugins
hot-reload
cost
metrics
timeseries
tracing-ui
replay-video
self-improve
eval
audit-export
anchor
atomic
locks
bus
stream
mcp
sub-reflex
skysend
explain
health
compress
workflow
abdeploy
external-api
registry
// Use cases

Built for ops that cannot afford silent failure.

🏠

Real estate

Buyer matching, property scrapes, agent notifications. Past failures: silent dedup loss, wrong-date matches. Fixed by ledger + verifier.

🛡️

Compliance / AML

KYC pipelines with audit trail. Every screen, score, decision signed + replayable for regulators.

💬

Customer ops

WA / email triage with approval gates. Refunds, escalations, sensitive replies require council vote.

📈

Sales automation

Cold sequences with hard idempotency. No double-sends. STOP respected. Rate limits ledgered.

🚀

DevOps

Deploys with snapshot + rollback. Incidents auto-open channels, fetch state, draft postmortem.

📰

Content ops

Publishing pipelines with multi-format fan-out, dedup, approval gates, full revision history.

Run agents like production systems, not prompts.

ZiggyOS ships embedded with ClawPipe. Or run standalone — your agents emit, ZiggyOS records.

Request access →