Cross-run memory
Every session is a transcript. Sibling agents on the same account list and peek each other's history — "what did the cron do yesterday?" becomes one API call.
01 — The trap
You're building the agent's backend by hand.
Where does it run? What does it remember? How does it get a secret without you pasting one in? Every serious agent ends up needing a session store, a sandbox, secret plumbing, and model glue — and you build all four yourself. cerver gives the agent those primitives behind one auth header.
02 — What the agent gets, on its own
Three primitives. One auth header. No human in the loop.
Compute on demand
The sandbox is a tool the agent calls. Swap compute mid-run, or spawn sub-agents — each on its own machine, each on its own harness (Claude Code, Codex, Grok).
Scoped secrets
Vaults per app, injected at run — the agent uses a token without ever seeing or storing it. Bring your own Infisical, or use cerver's.
03 — One call to run
Create a session, run it anywhere, peek it later.
Three calls and your agent has a place to think, act, and remember — model-agnostic and multi-tenant out of the box.
POST /v2/sessions → { id, transcript } create the agent's session
POST /v2/sessions/:id/run-llm model: claude/opus → grok hot-swap mid-run, thread intact
GET /v2/sessions/:id?tail=20 ← what a sibling did read any session's memory04 — At scale
Run thousands of agents in parallel — and see the cost of each.
Fan out a fleet, keep every transcript bound to its model and compute, and watch the cost per agent in the dashboard. Find the one that's looping, route the cheap work to a cheaper model, and prove the fleet is actually paying off.
05 — Install
One line. Point your agent at it.
$
curl -fsSL https://cerver.ai/install.sh | bash
Bring your own compute. cerver charges nothing — your bill is whatever your provider charges, usually $0 at hobby scale.