Wallets and keys
Vexthium works with two objects. Wallets own keys, and keys identify callers. Your team is the set of keys you choose to manage together; there are no organizations, seats or logins to set up.
That keeps trust simple. A key reads, simulates, prepares and builds inside its scopes, and signing stays with a wallet. Whoever holds a copied key can use its rate limit and read public chain data, and funds stay where they are.
A key for each agent
Issue a separate key to every agent, environment and integration. One wallet keeps five active keys; add another wallet for more.
| Situation | Setup | Why |
|---|---|---|
| The whole team calls Vexthium | A key for each agent | Usage and errors trace to one agent, and revoking one key leaves the others running. |
| Staging and production both run | One key per environment | A runaway test loop stays inside its own budget. |
| A browser app needs chain data | Open access, or a key kept on your backend | Code shipped to a browser is readable by anyone. |
Name each key for its job, such as support-bot-prod. That name shows up in your usage tables and, for public profiles, in the directory.
The owning wallet creates keys on the Agent keys page. The full key appears one time, so put it in your secret manager right then.
Narrow scopes
| Scope | What it covers | Suits |
|---|---|---|
read | Chain reads | Monitors, dashboards, support bots and research agents. |
simulate | Decoding, simulation and analysis | Agents that check or explain transactions for people. |
prepare | Transactions prepared for approval | Agents that hand people a transaction to approve. |
build | Contract templates, compiling and verification | Agents that write, compile and verify Solidity for a wallet. |
An agent that answers questions needs read, and simulate at most. The client never sees tools outside a key's scopes, so a prompt has nothing to point the model at.
Choosing the signer
Settle this for each agent before it touches mainnet.
A person approves
The agent prepares a transaction and returns an approval link. A person opens it, reads the new simulation and signs in their own wallet. This is the default, and it fits any action a human should sign off on.
The agent approves
The agent keeps its own key, signs locally and broadcasts through send_signed_transaction. Vexthium sees the signed result and the key stays with the agent. For this setup:
- Fund the agent wallet with the amount you are prepared to commit to it.
- Stop automatically on every
dangerflag, and on eachwarningyour policy excludes. - Keep an allow-list of contracts and functions in your own code; that list belongs to your agent.
- Store the signing key in a secret manager or a signing service, apart from prompts and repositories.
Usage data
On the Agent keys page the owning wallet sees each key: calls today, calls and errors across 30 days, a per-tool breakdown and the latest calls. An agent reads its own figures with its key:
curl -s https://vexthium.com/api/keys/me -H 'authorization: Bearer vxm_…'
Vexthium records each call's tool name, heavy or light, chain ID, result, error code and duration. Arguments and results stay out of the log, so usage data holds none of the addresses or amounts your agents handled.
Each key gets 600 calls and 60 heavy calls a minute. Service-wide totals are on the Stats page.
Public profiles
A key can hold a profile with a name, a description and a capability list. Make it public and /api/agents lists those three fields; the key, its hash and the owner wallet stay private. Keep profiles private until you want others to find the agent.
Exposed key playbook
- 01Revoke. Use the Agent keys page, or call
DELETE /api/keys/mewith the key from anywhere. It takes effect at once. - 02Replace. Create a new key with matching scopes and roll it out.
- 03Review. Read the usage of the old key and look for calls you cannot place.
- 04Check the signer. Agent keys carry no signing power. If the agent's own wallet key was stored next to it, review that wallet too.
Mainnet rollout
- 01Run the complete flow on the development network with
chainId: 46630, covering prepare, sign and wait, plus verification for contracts. - 02Run it again on mainnet, the default, with the smallest amounts that show the path works.
- 03Perps live on mainnet only, so try them there with small amounts.
- 04Switch on agent signing last, once person-approved runs have gone smoothly for a while.
Where each control lives
- Access runs through wallets and keys. Roles, seats and single sign-on stay with your own tooling.
- Funds and keys stay in your wallets. Vexthium prepares, simulates and relays.
- Spend caps and contract allow-lists for agent wallets belong in your agent or your wallet.
- Risk flags are structural rules, each one listed in the reference.