Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.usecompassai.com/llms.txt

Use this file to discover all available pages before exploring further.

Compass operates inside three rules attached to your smart account: a risk band, a protocol whitelist, and a chain whitelist. All three are stored on-chain and checked before any route executes. The agent has no authority to widen them — only your owner key can.
This page covers what each rule does, how to set it, and when changes take effect.

The three rules

RuleWhat it boundsSet by
Risk bandThe highest-risk venue the agent may useOwner
Protocol whitelistThe set of protocols the agent may callOwner
Chain whitelistThe set of chains the agent may move USDC toOwner
Every plan the chat agent generates is checked against all three before execution. A plan that fails any check is rejected and never broadcast.

Risk band (1–10)

A single integer from 1 (most conservative) to 10 (most aggressive). Each supported protocol is assigned a risk score; the agent may only route to protocols at or below your band. The mapping from band to protocols is published in Supported chains & protocols. This page is the source of truth — Compass does not list APR estimates in docs because real yields move daily and testnet yields are not meaningful.
Risk scores are assigned manually as part of venue review. New protocols cannot be added by the agent and cannot self-assign their score. See Authority & upgrade model.

Protocol whitelist

A list of protocols the agent is allowed to call. The list is a strict subset of Supported chains & protocols.
  • The agent can route to any whitelisted protocol that also satisfies your risk band.
  • The agent cannot route to a non-whitelisted protocol, even if you ask it to in chat — the policy check runs after the LLM generates the plan.
  • You can whitelist a protocol on one chain and not on another. The whitelist is per (protocol, chain) pair.

Chain whitelist

A list of chains the agent may move USDC to via Circle Gateway. Same model as the protocol whitelist:
  • Arc Testnet is always whitelisted — it’s your home chain.
  • Adding a chain to the whitelist does not automatically whitelist protocols on it. You whitelist each (protocol, chain) pair explicitly.

Per-route and daily caps

In addition to the three whitelists, your policy includes spending limits:
  • Per-route cap — the maximum USDC amount a single route can move.
  • Daily cap — the maximum USDC amount the agent can move across all routes in a 24-hour window.
Both caps default to conservative values on a new account. You can raise or lower them in the same UI as the whitelists.

How to set or change your rules

  1. Open the app and click Rules in the sidebar.
  2. Adjust the risk slider, toggle protocols and chains, set caps.
  3. Click Save rules.
  4. Your owner wallet will prompt for one signature to commit the new policy on-chain.
Once the transaction confirms (sub-second on Arc), the new rules are immediately active. The agent re-evaluates any in-flight plan against the new policy before the next tick.
Rule changes are owner-only. The agent’s session key has no permission to modify any of the rules above. See Session keys.

What happens if a rule is broken

Rules are checked at the contract level on every route. If the agent ever attempts a call that violates your policy:
  • The call reverts before USDC moves.
  • The attempt is logged to the audit trail with the failed check.
  • The agent does not retry the same plan — it generates a new one or pauses.
The agent cannot “exceed” your rules. The worst-case outcome is that it proposes a plan you reject, or it pauses because no valid plan exists inside your current rules.

Tightening rules during a route

You can change rules at any time, including while the agent has an open position:
  • Tightening the whitelist or lowering risk band — existing positions stay where they are, but the agent will not open new positions in venues that no longer satisfy the rules. You can Withdraw All to exit existing positions manually.
  • Loosening rules — the agent immediately becomes eligible to route to newly whitelisted venues on its next tick.

Next steps

Supported chains & protocols

The full list of what you can whitelist today.

Policy engine

How rules are encoded and enforced on every tick.

Session keys

The on-chain layer that makes “agent can’t break rules” true.

Audit trail

Every rule check, pass or fail, logged.