Access tiers
This page describes what exists right now, not a price list. There are no paid plans, no card forms and no service-level promises. The limits below are the ones the code enforces.
Accounts are managed at chainlayertwo.online/account. The anonymous and local tiers work without an account.
Install and go. No sign-up.
How you get itRun the one-line installer. It writes a shared public key into your CLI config for you.
Everyone on the shared key draws from the same gateway capacity. Busy moments can return 503.
ApprovalNone
Where requests goThe hosted gateway, directly
CostFree
Your own identity, history and a daily quota.
How you get itRegister, then chainlayertwo login to bind the CLI. New accounts start as pending.
Requests per account, reset at midnight UTC. Role user.
ApprovalNeeded. An admin approves the account before /v1 answers it.
API keysOne self-serve gateway key at a time (30 requests/min burst), created on the accounts panel
Where requests goThe hosted gateway, through the account service
CostFree
For people running real workloads with us.
How you get itAn admin grants the privileged (or admin) role to an approved account.
Still subject to what the gateway and its providers can serve.
ApprovalGranted by an admin, not requested in the app
API keysUp to five self-serve gateway keys, no per-key cap
Where requests goThe hosted gateway, through the account service
CostFree
Your models, your machine, no one else in the loop.
How you get itRun Ollama, llama.cpp or LM Studio, then chainlayertwo --provider ollama.
No quota. Speed and context size depend on your GPU and the model.
ApprovalNone
Where requests goYour own server on 127.0.0.1. Prompts never leave the machine.
CostFree
Not built yet.
How you get itYou can't yet. Nothing here takes payment.
ApprovalTo be decided
Where requests goTo be decided
CostNo price has been set
Your own provider keys work in every tier. Set DEEPSEEK_API_KEY, MOONSHOT_API_KEY or ANTHROPIC_API_KEY and the CLI can talk to that provider directly. Those requests go to the provider and are billed by them, not counted against any quota here.
Tools are separate. The MCP tool gateway always needs its own bearer token, whatever tier you are on. See ChainLayer2 MCP.
No. The installer configures a shared public key, so the CLI can reach the hosted gateway on first run. An account adds session history, the dashboard, and a quota that belongs to you instead of a shared pool.
Everyone using the public key goes through the same gateway, with no per-person allowance. When it is busy, a request can get 503 with a Retry-After header. The CLI waits and retries on its own. There is no guaranteed throughput.
Per account, per UTC day, on requests to /v1 made with that account's token. The count resets at midnight UTC. A request that never reached an upstream model (a timeout on our side, for example) is not counted. Once you hit the limit, /v1 answers 429 until the reset.
New accounts start as pending. A pending account can sign in and approve CLI devices straight away, but /v1 refuses its requests with an "awaiting approval" message until an admin approves it. That keeps the hosted gateway from being drained by throwaway accounts.
Ask the operator. The role is granted by an admin from the account panel and takes effect at once. There is no self-serve upgrade and nothing to buy. The admin role also has no daily cap, and it adds the admin tools.
There is no quota, because nothing passes through us: the CLI talks straight to Ollama, llama.cpp or LM Studio on your machine. The real limits are your hardware and the model you load. chainlayertwo providers local lists running servers and suggests a llama-server command sized for your GPU.
Maybe. The Pro column is a placeholder so this table doesn't change shape later. There is no price, no waitlist and no payment flow. When that changes, it will be on this page.
No. The hosted gateway is best effort in every tier. Gateway stats shows measured uptime and traffic, which is a record of what happened, not a promise. If you need something you control, use the local tier or self-host the stack.