Agent sandbox
Drag an agent onto a provider, or pick a route on its card. Each request is redacted with the daemon’s own rules, checked in the enclave, priced at the gateway’s live prices and held against its agent’s daily cap. The traffic is simulated: nothing leaves this page.
Reading the gateway’s prices…
Your agents
research-bot
Reads papers, drafts summaries
Sending to Claude
treasury-agent
Reconciles invoices
Sending to GPT
inbox-triage
Sorts mail, drafts replies
Idle
TEE enclave
Intel TDX
- Signature
- Receipt
- Escrow
- Metadata stripped
No gas per request: usage settles in batches.
Providers
Claude
api.anthropic.com
p50 186 ms
GPT
api.openai.com
p50 171 ms
OpenRouter
openrouter.ai
Llama 3.3 70B
p50 238 ms
- Requests served
- 0Each one signed and checked, none with gas
- Spent
- $00 ETH
- Saved against list prices
- $0Evane adds about 12 ms a request
- Next settlement
- 0 of 100receipts, or 5 minutes after the first
Requests
Waiting for prices…
What left research-bot’s machine
Its agent wrote:
Summarize the attached fraud report for removed: priya@example.com and list every charge on card removed: 4111 1111 1111 1111.
The daemon sent:
Summarize the attached fraud report for [email] and list every charge on card [card].
Same rules as evane-clientd. The provider sees only this, from the gateway, with Evane’s key.
To send real requests, fund an account with Connect Wallet and run evane-clientd: the quickstart takes a few minutes. What each model costs is on the pricing page.