For developers · onboarding in preview

Host the agent you already built.

You wrote the agent — the loop, the prompts, the model, the tools. plori gives it a computer to live on: a persistent disk, real tools, scale-to-zero when idle, isolation, and billing. You keep your code and change almost nothing.

See the manifest

Paste it into Claude Code, Cursor, Codex — whatever you build in.

You write coding agents. Let one do the integration.

§01

There's nothing to read and no SDK to learn. The whole integration is one brief: what plori is, the manifest schema, the endpoint contract, and worked examples. Hand it to your agent, review the diff, ship.

Copy the brief

One self-contained file — the manifest schema, the endpoint contract, and two worked examples. No docs site to crawl.

Paste it to your agent

Point it at your repo. It writes your agent-brain.yaml and adapts your server to the contract.

Review & ship

You read the diff. No new framework, no SDK — usually just a manifest and a health route.

or give your agent the URLplori.ai/build/agent-spec.md

The whole thing is one file.

§02

You onboard by declaring an AgentBrain manifest — it reads like a Kubernetes object. It says how to invoke your agent, how it stays healthy, what state to keep, and how it reaches tools. plori runs no agent loop and injects no tools; it owns everything around your loop.

agent-brain.yaml

apiVersion: agents.plori.dev/v1
kind: AgentBrain
metadata:
  name: my-agent
  vendor: acme
spec:
  image: ghcr.io/acme/my-agent:latest   # your container — the unit of pluggability

  invocation:
    profile: openai-chat                # openai-chat | openai-responses | ari-native
    endpoint: /v1/chat/completions      # the path we POST a turn to

  health:
    liveness: /v1/health

  lifecycle:
    model: one-shot                     # one-shot | persistent | session-affinity
    scaleToZero: true                   # idle costs nothing

  state:                                # externalize state here and you're stateless-on-host
    workingDir: /workspace              # we mount your agent's persistent disk
    sessionIdentity:
      kind: header
      name: X-My-Session-Id

  tools:
    protocol: skills                    # cli-skills on PATH + a SKILL.md; your shell runs them

The one real decision is invocation.profile — pick it by the HTTP surface your agent already has:

openai-chat

Already speak the OpenAI Chat or Responses API? Set the profile and you're basically done — we normalize the rest.

ari-native

Want the UI to show reasoning, steps, and tool calls? Emit our canonical events on /invocations.

adapter

Bespoke wire? Ship a thin sidecar that presents one of the above. The escape hatch for anything.

You keep your code. We handle the computer.

§03

You keep

Your agent loop, prompts, and reasoning
Your choice of models and providers
Your framework — any language, any design
Your tools (CLI skills on your sandbox)

We handle

Hosting & one-shot scale-to-zero (idle costs nothing)
A persistent per-agent disk across runs
Multi-tenancy, isolation, and secrets
Tracing, usage, and billing

Self-serve onboarding is on the way.

§04

The interface is real and live in the runtime; the public upload flow is still being built. Tell us what you're building and we'll bring you in as soon as it opens — early builders help shape the contract.

Prefer to talk first? Email dev@plori.ai.