Compute Sandboxes · run9 Agent runtime · smith9 · soon Browser · web9 · soon Data Postgres database · db9 Object storage · drive9 Agent memory · mem9 Secrets · vault9 · soon Intelligence Model gateway · gate9 · soon Skills & tools · hub9 · soon Evals · eval9 · soon Coordinate Queues · task9 Realtime · inbox9 · pulse9 · tape9 Teamwork · chord9 · soon Scheduling · cron9 · soon Operate Observability · owl9 Auth · auth9 · soon All products → Explore Solutions Pricing Customers Enterprise CLI Docs Company GitHub Request access
Solutions · Switch from DIY

You already built this. Stop maintaining it.

A working agent stack is usually five vendors in a trench coat: a microVM service, object storage, a queue, a vector DB, and some auth you wrote. It works — and every piece is yours to patch, bill, and page on. sys9 replaces it with atomic, agent-native services you compose, and opt into one at a time.

the swap

What you wired together → one service.

A part-for-part map from the stack you maintain to the service that replaces it. Take the rows you need; leave the rest.

What you wired togetherThe sys9 serviceWhat changes
microVM vendor (e.g. E2B, Firecracker)run9Fork the whole OS — data and packages — in ~50ms. Run up to 1,000,000 in parallel, pay only when running. No image pipeline to maintain.
Object storage (e.g. S3) + gluedb9 + drive9File storage built into the database, no bucket wiring — plus a shared FUSE filesystem across sandboxes with semantic grep.
A queue / job runnertask9 + tape9Atomic work-claim with no queue to declare, and an append-only stream with real-time replay. One winner per task, liveness-backed.
A vector DB / search servicedrive9 + mem9Semantic grep (vector + BM25 in parallel) over your files, and one persistent memory shared across every agent.
Agent-to-agent messaging glueinbox9 + pulse9FIFO mailboxes with guaranteed delivery and a liveness registry — the coordination layer, without the broker you stood up.
Homegrown authauth9 · soonSign in once, unlock every service; use-first claim-later for your service too. Until it lands, scope secrets per agent with drive9’s vault.
DIY logging / tracingowl9Wrap the binary for zero-config capture — sub-agents auto-traced, human-readable transcripts, no code changes.
assembled vs composed

Same capabilities. Far fewer seams.

A DIY stack

Capabilities glued across five vendors you operate yourself.

  • Five bills, five dashboards, five on-call surfaces
  • Glue code between every pair of services
  • Idle compute often still costs you
  • Auth and isolation are yours to get right
  • Not built for agents — you adapt SDKs to the loop

sys9

Atomic services that compose, driven by APIs, CLIs, and Skills.

  • One agent-native suite; opt into only what you use
  • Services designed to compose — less glue to write
  • Pay only when running; idle sandboxes sleep free
  • Isolation by construction — a fork per tenant, not a policy
  • Use-first, claim-later — hand an agent the skill
how to move

Migrate a row at a time.

You don’t have to rip out the stack. Replace the piece that hurts most, keep the rest, and compose from there.

01

Start with the sharp edge

Swap the one service that pages you most — usually compute. run9 forks drop in where your microVM vendor sat.

02

Compose outward

Add storage, a database, coordination, or traces as you go. Each is independent; nothing forces a big-bang cutover.

03

Retire the glue

As services move over, the integration code between them goes away — the seams were the point.

use-first, claim-later

Replace the stack one service at a time.

Tell us what your DIY stack looks like. Start by forking a sandbox today — no signup, pay only when it runs.