# plori > plori gives you a cloud AI agent with its own computer: a persistent disk, > real CLI tools, and memory. Asleep, an agent stops billing for compute. > Billing is pegged at 120 credits per US dollar (1 credit ≈ $0.01). plori gives you an AI agent on its own computer. Each agent keeps a persistent disk, real CLI tools, per-agent isolation, and memory; asleep, an agent stops billing for compute. Describe a routine once and the agent builds it into a standing workflow that fires on a schedule or a webhook without an agent turn. Use it from the web, or let your own agent drive it from any MCP client. Reach for it when the work should outlive a single request: files written now and read next week, software installed once and reused, a repo kept checked out, a long job handed off and read later, or a recurring routine turned into an AI-built workflow. For a stateless snippet with no state worth keeping, a plain code sandbox is cheaper. No signup needed to try. ## Start here: one path - The no-signup trial includes 100 credits, 128 MB of disk, and is kept for 7 days. - Add plori to Claude Code: `claude mcp add --transport http plori https://api.plori.ai/mcp`. - On first use, the MCP client opens Plori's email-code OAuth sign-in; no API key is needed. - First task: "Create a cloud agent on plori, send it a hello message, and show me its reply." - Full agent-facing briefing: . ## Core pages - [Home](https://plori.ai/): What "its own computer" means, pricing, and driving plori from your own agent. - [Pricing & FAQ](https://plori.ai/pricing): Usage-based credit rates, worked cost examples, and answers to common billing questions. - [MCP: drive plori from your agent](https://plori.ai/mcp): Connect any MCP client to api.plori.ai/mcp (OAuth sign-in, or an API key for headless use); create and run agents, read replies, work the human-in-the-loop queue, schedule runs, and build and run workflows; plus the retry (idempotency key) and per-plan concurrency semantics a machine caller needs. ## Docs - [Docs index](https://plori.ai/docs): All guides. - [Quickstart — start here](https://plori.ai/docs/getting-started): Start free, give the agent a real task, connect a terminal, and pay only when the free allowance no longer fits. - [plori CLI](https://plori.ai/docs/cli): Install, browser login, first run, JSON output, CI API keys, and credit links. - [Disk and files](https://plori.ai/docs/disk-and-files): One account disk pool all your agents draw from (1/20/50 GB by plan, +10 credits/GB-month), the files panel, and what persists. - [Publish a site](https://plori.ai/docs/publish-a-site): Serve one directory of an agent's disk at .plori.site; private by default with a public toggle, and it keeps serving while the agent sleeps. - [Agent memory](https://plori.ai/docs/memory): Notes the agent reads every turn; private vs account scope; the memory panel. - [Scheduled runs](https://plori.ai/docs/scheduled-runs): Deferred one-shot runs via chat or the schedule_run MCP tool; how replies are delivered. - [Background tasks](https://plori.ai/docs/background-tasks): Long-running commands that keep going after the reply and post results back in-thread. - [Workflows](https://plori.ai/docs/workflows): Standing automations the agent builds in chat; schedule/webhook/manual triggers, safe test runs, sealed credentials, 1 credit per run with 90 seconds of runtime included, then 1 credit per started minute (24-hour safety stop), plus model usage for AI steps at each model's own rate. - [Bring your own model key](https://plori.ai/docs/byok): OpenRouter one-click OAuth or pasted keys (OpenAI, DeepSeek, Groq, xAI, Together); model usage stops drawing credits. ## Solutions - [AI agent hosting](https://plori.ai/ai-agent-hosting): What hosting an agent on plori includes and costs. - [A cloud computer for AI agents](https://plori.ai/cloud-computer-for-ai-agents): The persistent machine model vs ephemeral sandboxes. - [Background agents](https://plori.ai/background-agents): Agents that work unattended: background tasks, schedules, workflows, human-in-the-loop. - [AI-built workflows](https://plori.ai/workflows): Standing automations your agent builds from a described routine; schedule or webhook triggers, safe test runs, 1 credit per run, and creatable from any MCP client. - [Run a coding agent in the cloud](https://plori.ai/cloud-coding-agent): Repos, builds, and tests on the agent's own machine. - [plori vs alternatives](https://plori.ai/alternatives): Honest comparison with sandbox APIs, a VPS, local, and IDE agents, including when NOT to use plori. ## Benchmarks - [Agent cold-start latency](https://plori.ai/benchmarks/agent-cold-start): How fast a sleeping agent wakes, measured end to end against production; full methodology plus downloadable raw data (CSV/JSON, CC BY 4.0). ## Engineering blog - [Blog index](https://plori.ai/blog): Engineering notes from building plori; measured numbers and production incidents. - [What a Kubernetes warm pool cannot prewarm for a stateful AI agent](https://plori.ai/blog/what-kubernetes-warm-pool-cannot-prewarm): The boundary between tenant-agnostic prewarming and atomic claim-time binding of an agent's disk, identity, and credentials. - [The workflow engine said OK after a failed step](https://plori.ai/blog/workflow-engine-ack-not-success): Why a transport acknowledgement is not workflow success, and how a durable terminal record becomes the authoritative execution result. - [Why we put a server between our AI agents and JuiceFS](https://plori.ai/blog/juicefs-multi-tenant-isolation): Where JuiceFS CE's trust boundary sits (the client) and the mTLS server plori built in front for per-agent isolation. - [443 dead FUSE mounts wedged kubelet: a production postmortem](https://plori.ai/blog/dead-fuse-mounts-kubelet-postmortem): Why kubelet cannot tear down a dead FUSE mount (ENOTCONN), the three diagnosis traps, and the 60-second janitor that reaps them. - [JuiceFS metadata: Postgres to Redis, ~7x faster](https://plori.ai/blog/juicefs-metadata-postgres-to-redis): A production metadata-engine migration with the head-to-head numbers, the durability config, and the dump/load procedure. ## For agents and MCP clients - [Integration front door](https://plori.ai/agents.md): start here. Routes an agent to MCP, REST, or the docs by intent, with a first task to run. - plori's MCP server is at `https://api.plori.ai/mcp` (Streamable HTTP). Auth: the standard MCP OAuth handshake (a 401 starts it; dynamic client registration is open), or `Authorization: Bearer `. Per-client connect commands and the tool list: . Credential walkthrough: . - Skill catalog: (also at /.well-known/agent-skills/index.json). ## Markdown for agents - Every public page above (including all docs and solutions pages) is served as Markdown two ways: append `.md` to its path (home is `/index.md`), or request the page URL with the `Accept: text/markdown` header. The response sets `Content-Type: text/markdown` and an `x-markdown-tokens` estimate. ## Pricing - Awake compute: registered accounts include 2,000 minutes of run time each month; past that it is 1 credit / minute. Anonymous trials bill run time from the first minute. Idle agents sleep and stop billing for compute. Model usage always draws credits. - Disk: plan-included disk is free; extra disk is 10 credits / GB / month (disk is the only cost while asleep). - Model usage: billed at the model's real provider cost, converted at 120 credits per USD (each call rounds up to a whole credit). The Plori Router picks the cheapest model that fits each task; paid plans add stronger models. Bring your own key and model usage is 0 credits. - 1 credit ≈ $0.01. No signup needed to try. ## Contact - Developer onboarding & partnerships: agent@plori.ai