$ANSEM received7xK…9fQ
Agent balance$5.00
Pay in $ANSEM
Send $ANSEM to fund your agent's balance. We convert it the moment it lands, so price swings never touch your balance.
Your agent is overpaying to think.
Let it graze.
Pay for AI compute in $ANSEM. Graze sends every call to the cheapest model that can do the job.
Get a keySame answers. A fraction of the bill.
93%cheaper per typical call than Claude Sonnet 4.6 at list price
$0.49for 1,000 typical calls, down from $7.40
5%Graze's fee, shown on every call
1setting to change in your agent
Measured on real calls on 25 September: about 1,700 tokens in and 160 out, on llama-4-maverick, the first model graze/auto tries. Costs are what UsePod actually charged, plus Graze's 5% fee.
How it works
Three steps. None of them are yours.
Fund once, point your agent at Graze, and every call finds its own cheapest route.
llama-4-maverick
tried firstAnswered
tried firstAnswered
gpt-6-luna
if neededStandby
if neededStandby
Cheapest first
Every call tries a fast, cheap model first. If it errors, or the answer comes back empty or broken, the next model takes over.
X-Graze-Model llama-4-maverick
X-Graze-Attempts 1
X-Graze-Cost-USD 0.000486
X-Graze-Saved-USD 0.006912
Every call, itemised
Each response tells your agent what it paid and what it saved. No monthly surprises.
← swipe for the rest
Live
Every blade is a real request.
The dashboard is public. Watch agents graze in real time.
The field is empty. Waiting for the first agent.
Calls0
Tokens0
Paid$0
Saved$0
No demo data
Every number comes from a real call. The window above is reading them live.
Open to anyone
Judges, holders and other builders all see the same page.
Built for agents
Plugs in anywhere. Never overspends.
base_url https://graze…/v1
model graze/auto
Save
One URL change
Works with any agent that speaks the OpenAI format. No SDK to install.
llama-4-maverick
Error 503
gpt-6-luna answered
● Done
Failover built in
If a model errors, the next one answers. The failed attempt costs nothing.
Balance
Spend can't run away
The worst-case cost is set aside before every call, so a balance never goes below zero.
Models from every lab
Four today, from DeepSeek, Moonshot, OpenAI and Meta. The UsePod marketplace has more, and adding one is a line in a price list.
DSKOAMQG
ZXMMA+
Savings you can chart
Every call is logged with what it paid and what it saved, and the live field adds them up.
The token
$GRAZE. Every call feeds it.
Agents pay for compute in $ANSEM. $GRAZE ties the token to how much Graze gets used.
Contract address
Launching soon
Launching on ClawPump before 1 October. The address appears here the moment it's live, no refresh needed.
Only trust the address shown here or posted on our X. Check it before you buy.
Agents pay the 5% fee
A share buys $GRAZE
Fee buyback
A share of Graze's 5% fee buys $GRAZE on the open market. The more agents graze, the more it buys.
Holders stake $GRAZE
New agents get starter credit
Staking for starter credit
Holders stake $GRAZE to fund starter credit for new agents, so builders can try Graze before they buy anything.
Honest
What Graze doesn't do yet.
$ANSEM deposits
Being built now. Until they ship, balances are topped up by hand.
Answer quality
The check catches empty answers, broken JSON and missing tool calls. It doesn't judge whether an answer is good.
Streaming
A streamed reply can switch models only before it starts. After that, it isn't checked.
Custody
Graze holds your balance, so this is a beta. Balances are capped at $50 per agent.
Prices
Graze bills what UsePod actually charges for each call, plus 5%. UsePod's prices can move, and so can the savings.
Token utility
Buyback and staking are planned, not live. The share of fees and the staking terms aren't set yet.
Hermes plugin
In progress. For now, point Hermes at Graze's URL.