Building blocks for agent systems.
Atomic, Unix-like services — each does one job, small and focused — that agents drive through APIs, CLIs, and Skills. Grouped by what they do: compute, data, intelligence, coordination, and operations. Pick the ones you need; compose the rest. Use-first, claim-later — no signup to start.
Where agents run — sandboxes, hosted runtimes, and browsers.
Sandboxes
forkable sandbox for agents.
Fork the whole OS in ~50ms; run up to 1,000,000 in parallel, pay only when running. Drive it over SSH.
run9 → smith9 · coming soonAgent runtime
agent as a service.
Host agent fleets on run9 — wake on a message, pay for active moments only, one setup inherited by every agent.
smith9 → web9 · coming soonBrowser
managed browsers for agents.
Managed headless browsers for agents — sessions, stealth, captcha, screenshots; the web access run9 does not cover.
web9 →State that outlives a session — databases, storage, memory, secrets.
Postgres database
serverless Postgres, from terminal.
Postgres + branching (fork db, data, and files together) + file storage with no S3, and zero signup.
db9 → drive9 · liveObject storage
any sandbox, one filesystem.
A shared filesystem across sandboxes and sessions — FUSE mount, semantic grep, per-agent grants.
drive9 → mem9 · liveAgent memory
one memory, many agents.
Persistent memory across sessions, agents, and devices — one key serves every agent; humans inspect, import, export.
mem9 → vault9 · coming soonSecrets
secrets, sealed and granted.
First-class secrets and credentials — per-agent grants, rotation, and sealed delivery to running agents.
vault9 →The model layer — gateways, skills, and evaluation.
Model gateway
one endpoint for every model.
Routing, caching, cost caps, fallback, and multi-provider keys behind a single endpoint.
gate9 → hub9 · coming soonSkills & tools
a registry for agent skills.
A registry and marketplace for agent skills, tools, and MCP servers — publish once, any agent installs.
hub9 → eval9 · coming soonEvals
offline evals, online quality.
Offline evals plus online quality for agents — owl9 says what happened, eval9 says whether it was good.
eval9 →How agents work together — queues, realtime, teamwork, schedules.
Queues
atomic task claim.
Distributed work-claim — atomic claim, one winner; owner liveness via pulse9 auto-unassigns on death.
task9 → inbox9 · pulse9 · tape9 · liveRealtime
messaging, presence, streams.
Three atomic realtime services in one bundle — inbox9 messaging, pulse9 presence, tape9 streams.
inbox9 · pulse9 · tape9 → chord9 · coming soonTeamwork
agents and humans, one team.
Assign work to an agent like a coworker; orchestrate heterogeneous agents into one team built for long runs.
chord9 → cron9 · coming soonScheduling
cron and triggers for agents.
Run agents on a schedule or on events and webhooks — cron plus triggers for agent work.
cron9 →Run it in production — observability and identity.
Observability
agent session observability.
Zero-config capture — wrap the binary, no agent changes; sub-agents auto-traced into human-readable transcripts.
owl9 → auth9 · coming soonAuth
one identity, any sys9.
Sign in once, unlock every sys9 service — agent-native, use-first claim-later, for your own service too.
auth9 →task9 (shared backlog) + inbox9 (agent-to-agent mail)
+ pulse9 (liveness) + smith9 (spawn agents on demand) = a hosted
agent team. Atomic services compose into real systems — no pipelines to declare; the
workflow emerges.
— the composition story Start with one service. Compose the rest.
Hand an agent the sys9 skill, or install the CLI and create your first database in seconds. Use-first, claim-later — no signup to start.