Agent investment
Delegate a bounded investment strategy to an autonomous agent — non-custodial, capped on-chain, run from your Safe.
Not shipped. The DCA rail described in this section is a design, not a working feature. The Strategy tab is hidden from the Safe App — the rail needs the Permit2 setup the limit order has before it can execute — and these pages are kept out of the documentation sidebar for that reason.
For the agent strategies that do work today, see Agent DeFi: provide liquidity and buy the dip.
Agent investment would let a Safe delegate a recurring investment strategy — dollar-cost averaging (DCA) — to an autonomous agent, without ever handing over the keys or the treasury.
The design below is worth reading anyway: the two-safe architecture and the EIP-1271 owner-set constraint apply to every agent mandate, including the ones that ship today.
The Safe signs one mandate. An agent then runs the strategy on the Safe's behalf, bounded on-chain by a per-swap cap it can never exceed. The treasury stays in the Safe; the agent only triggers the swaps and pays gas.
Strategies are software. Guardrails are consensus.
How it fits together
| Layer | Who | What |
|---|---|---|
| The mandate | the Safe (multisig) | one signed delegation: what the agent may do, capped per swap |
| The strategy | the agent (off-chain) | when and how much to buy — the operator's instruction |
| The guardrail | the caveat (on-chain) | the hard limit the contract enforces on every swap |
Read next
- Strategies — what a strategy mandate is, and why it bounds the balance change rather than the calldata.
- Getting started — set up and run the agent that executes the mandate.
The agent side can be fully automated with the Hourglass agent skill, which guides an LLM agent through the whole flow.