The substrate under your AI product.
Whether you’re shipping an AI feature or running a data pipeline, you need the same four things: memory that persists, a database, compute that scales out, and files your steps can share. sys9 gives you each as an atomic service — compose them, don’t assemble a stack.
Four services, one AI product.
Take memory on its own, or compose all four into a full pipeline. Opt into only what your product needs.
One memory, many agents
Persistent memory that outlives resets, restarts, and devices. One key serves every agent; humans can inspect, import, and export it. Integrates with Claude Code, Codex, OpenCode, OpenClaw, Hermes, and Dify.
mem9 → db9Postgres + storage
Serverless Postgres with branching and built-in file storage — no S3, zero signup. Fork the database, its data, and its files together to test a migration or a prompt change against real data.
db9 → run9Compute that forks
Run pipelines on a forkable sandbox: shard a batch across a hundred parallel forks, sub-second start, pay only when running. Branch the whole stack to compare two approaches side by side.
run9 → drive9Files + semantic grep
A shared filesystem across every step and session, with semantic grep (vector + BM25 in parallel) for retrieval and a per-agent secrets vault for the keys your pipeline needs.
drive9 →Shard the work, share the memory.
Fork compute for the parallel parts, read and write a real database, recall one memory across every agent, and retrieve context semantically — all from the terminal or an agent skill.
Parallel where it pays. Persistent where it counts.
mem9 keeps one memory across every agent and session, so context isn’t re-derived on each run. db9 holds the structured state. run9 forks the heavy stages out to a hundred sandboxes. drive9 carries the files between them.
run9 fork pipeline --as batch-7 --shards 50 ✓ 50 sandboxes, sub-second start db9 query "select id from documents limit 1000" ✓ 1000 rows · serverless Postgres mem9 recall --key user-42 ✓ one memory, across every agent drive9 fs grep "Q3 pricing" / ✓ vector + BM25, in parallel
Composable, agent-native, no lock-in.
Atomic & composable
Each service does one job. Start with one, add the rest when you need them — there’s no bundle to buy or framework to adopt.
Use-first, claim-later
No signup to start. Hand an agent the skill and it provisions what it needs; attach an account when you’re ready to keep it.
Works with your agents
mem9 plugs into the runtimes you already use, so memory and retrieval aren’t tied to one agent framework.
The services behind this shape.
One memory
Persistent cross-agent, cross-session memory; humans can inspect it.
mem9 → db9Postgres + storage
Serverless Postgres + branching + file storage, zero signup.
db9 → run9Forkable compute
Fork the OS in ~50ms; shard across 100, pay only when running.
run9 → drive9One filesystem
Shared FUSE filesystem, semantic grep, per-agent secrets vault.
drive9 →Build your AI product on composable parts.
Hand an agent the sys9 skill, or install the CLI and start with the one service you need today.