Enterprise

Run agent fleets in production — with control, not guesswork.

sys9 is the agent-native stack for teams running many agents across many tenants. Isolation, pay-only-when-running compute, zero-config observability, and per-agent identity — small, focused services you compose and opt into, not a monolith you adopt.

Talk to us See how isolation works →

~50ms

Fork the whole sandbox — OS, data, packages (run9)

up to 1M

Sandboxes in parallel, pay only when running

Pay-per-run

Billed only while running — sleep on idle

Zero-config

Session capture, sub-agents auto-traced (owl9)

what your team gets

Production concerns, solved as primitives.

Each service does one job. Use the ones you need; the rest stay out of your agents' context and off your bill.

Isolation

Fork an entire sandbox per agent or per tenant in ~50ms with run9. Branch a database, its data, and its files with db9. Give each agent its own filesystem and scoped vault grants with drive9.

run9 · db9 · drive9

Cost control

run9 wakes on call and sleeps on idle, so you pay only for the moments an agent is actually running. smith9 hosts fleets on the same model — pay for active moments, not idle capacity.

run9 · smith9 (coming soon)

Observability

owl9 captures sessions with zero config — wrap the binary, no agent changes (OWL9_ENABLE=1 claude). Sub-agents are auto-traced. Transcripts are human-readable, consistent across runtimes.

owl9

Identity

auth9 gives one identity across every sys9 service — sign in once, unlock the stack. Agent-native by design: Skill, CLI, and self-service API, use-first and claim-later.

auth9 (coming soon)

Memory & coordination

mem9 keeps one persistent memory across sessions, agents, and devices. task9, inbox9, and pulse9 give fleets a shared backlog, ordered messaging, and liveness — the atoms of a real agent team.

mem9 · task9 · inbox9 · pulse9

Composable, opt-in

No lock-in monolith. Adopt run9 today, add owl9 tomorrow, layer db9 when you need it. Each brick is independent, with its own CLI, so you migrate one capability at a time.

the whole stack, à la carte

the recipe

Atomic primitives compose into a hosted agent team.

task9 (a shared backlog) + inbox9 (agent-to-agent mail) + pulse9 (liveness) + smith9 (spawn agents on demand) compose into a hosted Claude Code agent team. No pipelines to declare — workflow emerges from atoms that each do one job well.

task9 inbox9 pulse9 smith9

terminal

# shared backlog — one winner per claim
task9 create "Triage flaky deploy"
task9 claim 41

# agent-to-agent mail, FIFO per recipient
inbox9 send bob --payload '{"text":"on it"}'

# who is alive right now?
pulse9 list → 3 agents live

rollout

Adopt one brick at a time.

Start with compute

Move agent runs onto run9 sandboxes. Fork the whole stack for parallel work; pay only when running.

See everything

Turn on owl9 with one env var. Get human-readable transcripts and auto-traced sub-agents across every runtime.

Add data & memory

Branch per-tenant databases with db9, share a filesystem with drive9, and give agents lasting memory with mem9.

talk to us

Tell us what your fleet needs to do.

We work with teams shipping agent systems and per-tenant AI. No signup required to start — hand an agent the skill, or reach us directly.

Talk to us Review the security posture →