A separate API key and spending limit for every AI agent.
Cedra is building a model router for AI agents. You will top up one balance and call frontier and open-source models through one API. Each agent will get its own API key and spending limit, so one runaway agent cannot drain the balance the others need. When a provider fails, requests will switch over automatically.
- research-agent $12.40 / $50.00 Running
- support-agent $18.90 / $20.00 Near limit
- nightly-loop $5.00 / $5.00 Paused at limit
- support-agent provider-a/model-x failed, switching
- support-agent fallback provider-b/model-x OK
- nightly-loop reached $5.00 of $5.00, paused until reset
Decide what each agent can spend.
One runaway loop should not drain the whole budget overnight. The agent is the unit we are building around: its own key, its own limit, its own bill per request.
-
A key and a limit for every agent
PlannedEach agent will get its own key and spending limit. When an agent hits its limit, it stops spending, and the others keep running.
-
Pay for what you call
PlannedYou will top up one balance and pay per request from it. Each agent's limit keeps one runaway agent from using up the balance the rest of your agents need.
-
Aiming for models served by their own providers
In developmentOur goal: every model served directly by its provider, not a reseller. You will be able to pin a provider when you need consistent behavior from call to call.
-
Your agent keeps running
PlannedWhen a provider fails, requests will switch over automatically instead of stopping your agent halfway through a task.
A model can think. An agent has to act.
You will not need to touch a blockchain to use Cedra. When your agent needs tools or chain data, they will come from the same stack, on Cedra, our Move L1.
Tools without the glue code
Agents will get access to onchain and offchain tools without you writing glue for every service.
Chain data over GraphQL
Balances, NFTs and transaction history, already indexed and queryable over GraphQL.
Verifiable randomness
A built-in oracle on the Cedra L1, starting with verifiable randomness.
Talk to other chains
Send messages to other chains from the same stack.
Five integrations, one stack.
An agent that reasons, stays within budget, reads chain data and talks to other chains usually means separate accounts, SDKs and bills for each. We are building Cedra so it is one.
Without Cedra
- Model accessa router or several provider keys
- Spend limits per agenta key per agent on your router, mapped and tracked by hand
- Randomnessseparate oracle
- Chain dataseparate indexer
- Cross-chainseparate bridge protocol
With Cedra
- Model accessin development
- Agent with its own key and limitplanned
- Oracle, VRF randomnessbuilt into the chain
- Indexer, GraphQLbuilt into the chain
- Cross-chain messagingbuilt into the chain
What builders ask first.
Can I cap what an agent spends?
That is the plan for launch: each agent will get its own key with its own spending limit.
Will I need to top up a balance?
Yes. You will top up a balance and pay per request from it. Per-agent limits keep one agent from draining that balance for the others.
Will the models be the same as calling the provider directly?
That is our goal: every model served directly by its provider, not a reseller, with the option to pin a specific provider for a model.
What happens when a provider goes down?
Requests will switch over automatically, so your agent does not stall in the middle of a run.
How is this different from other model routers?
Most routers already let you cap spend per API key, and that works. Cedra is building around the agent instead of the key: each agent will get its own key and spending limit as one unit, so you do not map keys to agents by hand. Underneath is our own Move L1: chain data and cross-chain messaging already run there, and agent tools are in development.
Do I need to know anything about blockchains?
No. If you only need models, you will use Cedra like any model API. The onchain tools will be there when your agent needs them.
When can I use it?
Cedra is in development. Join the early access list and you will get a key when your wave opens. We will open access in waves.
What chain does it run on?
Cedra is its own Move-based L1 with parallel execution, designed for sub-second finality.
Get an early key. We invite in waves.
The router is in development. We will let developers in by waves, as many as we can onboard by hand. Tell us what hurts most and we will use it to decide who goes first.
- An API key before public launch
- Beta access to per-agent keys and spending limits as they ship
- A say in which models and providers we add first
What happens after you sign up. We read your answers and place you in a wave. When your wave opens, you will get an email with your key and a quickstart. Waves are small on purpose, so each group gets real attention from the team. We do not have wave dates yet.