Skip to main content

Chat

Chat is where you build plans, on app.signo.fi. Home is the default dashboard. Type an action and Signo parses it into a structured intent (swap, bridge, deposit, withdraw, transfer, stake, claim, loop leverage, manage LPs, enter or exit Pendle PT/YT positions, or a chain of these), pulls live data (prices, balances, routes, yields), and returns a step-by-step plan you sign in your wallet. Related work is grouped into named conversations so you can return to it later.

Conversations​

  • Start a new conversation for a task or position.
  • The first prompt supplies an automatic title; you can rename it.
  • Plans and Agent results stay in the transcript with their original status.
  • Rebuild a stored plan when you need fresh quotes before signing.
  • Older runs load as you scroll, within your plan's history window: 30 days on Free, 90 on Plus, and no plan-imposed window on Pro. Older history is hidden, not deleted because of the window, and becomes visible again on a plan that covers it. See Pricing.

For a signed-in request using one of your wallets, closing the tab does not cancel the server's work. Return to the conversation to read its saved answer or failure status. This does not sign or execute a transaction for you.

Deleting a conversation removes it from the Chat list. It does not delete the underlying runs. Use the account-level privacy and deletion controls in Settings when you want to remove stored account data.

Advanced reasoning with your subscription​

On any plan, Settings → AI Models → Advanced reasoning can use models from a subscription you already have. On Signo Free those turns count toward the 15 requests a day. Only providers shown in Settings are available. The hosted connection supports ChatGPT today; Claude is not offered until its runtime is enabled.

Connecting ChatGPT opens the provider's device-login page. Starting the private login runtime can take up to a minute; Settings shows Connecting… while it starts. Signo stores the resulting login only in an isolated hosted runtime; its database stores the provider, connection status, selected model, and an opaque pointer rather than the login credential. The hosted login expires after 30 days unless use refreshes it. Disconnect asks the runtime to delete the login and its account-tagged copies immediately; any copy that cannot be deleted expires within 30 days.

When Advanced reasoning uses that connection, Signo sends the current prompt, relevant Chat history, and the instructions needed to answer to the selected provider under your account. Your provider's terms, privacy settings, availability, and usage limits apply. If the connected model cannot answer, Signo can fall back to its standard model and tells you.

Settings shows a green Enabled badge on the usable provider currently answering Chat and a gray Disabled badge on another usable connected provider. A pending, disconnected, or failed row has no status badge; its sentence and button show the available next step instead. Enable selects that provider and turns on Advanced reasoning. Disable switches Chat back to Signo's model but keeps the connection saved; Disconnect removes it as described above. The toggle and enabled provider are remembered in that browser.

When the connected model can answer, the first loading line after you send a message says Thinking with Advanced Reasoning (ChatGPT model)…. A later progress line can replace it as the turn continues. When Advanced reasoning is off or the connection cannot answer, Chat starts with the plain Thinking… line and Signo's standard model handles the turn.

Planning and signing​

Signo proposes a plan, supported routes are checked and refreshed, and your wallet signs each transaction. A saved conversation cannot sign or execute anything on its own.

Chat is not only a blank prompt box. It can show suggested prompts, suggested flows, and position-aware next steps based on wallets you have linked or scanned. Suggestions are starting points; you can edit them before running.

For questions about current public information, Chat can search the web and use those results in its answer. Web results are informational context, not a transaction instruction or financial recommendation.

For questions about how Signo works, Chat can answer from these public docs. You can also follow an Ask Signo link on an Agent notice to ask about that message.

What works​

  • Multi-chain. Ethereum, Base, Arbitrum, Optimism, Polygon, X Layer, Solana, and Robinhood Chain — all addressable in one prompt where the route is supported. See Protocol Coverage.
  • Multi-wallet. Reference wallets in your account by name or address, within your plan's wallet allowance. An EVM position read covers up to ten wallets per call; any wallets not read must be identified rather than treated as having no positions.
  • Slash shortcuts. Type / in the prompt to open the shortcut menu. Choosing an item inserts it into the prompt for you to keep editing or send. /wallet opens your wallets, /app opens protocols, /asset opens the token catalog, and /action opens supported actions.
  • Chained operations. "Bridge $X to Y then swap to Z" returns as a single plan; downstream amounts re-estimate after each step confirms. If the requested chain is short on an asset but another connected EVM chain holds enough, Signo may add a funding bridge to the plan. The bridge is shown as its own reviewable step before you sign.
  • Plain-English conditions. "If APY > 5% on Aave" or "only if my health factor stays above 1.5" — the planner respects natural-language guardrails alongside the structured intent.
  • Pendle PT, YT, and LP actions. Enter or exit PTs and YTs, redeem or roll a matured PT, and add or remove Pendle liquidity. Signo quotes the Pendle leg from live market data before it builds the action.
  • Uniswap pool discovery. Ask Chat to find a Uniswap pool by token, pair, or chain on Ethereum, Base, or Arbitrum. Results come from DefiLlama's public yields index, default to at least $1 million TVL, and sort by trailing fee APY unless you ask for TVL. Fee APY is historical, not a forecast. Only standard Uniswap v4 pools without hooks, with both tokens routable, can be opened; v3 pools are reference-only for a new position.
  • Aero actions on Base. For a swap, Signo compares verified direct classic stable and volatile pools, allowlisted Slipstream pools, and bounded two-hop classic routes through WETH or USDC, then uses the best quoted output. It refuses an Aero quote that would lose more than 15% of the input value and points you to an aggregator route instead. You can also add or remove classic liquidity, stake or unstake LP in a gauge, claim fees and AERO emissions, open or manage Slipstream concentrated positions, and create, top up, extend or withdraw timed veAERO locks. Signo verifies pools, gauges and position managers on-chain, uses native ETH directly where supported, and keeps each Aero action in its own plan.
  • Aero early-penalty consent. A newer Slipstream gauge can charge a penalty when you claim AERO or withdraw a staked position before its minimum stake time. Chat reads the live rate and end time and asks you to accept the quoted rate before it builds the action. After you choose Yes, Signo reads the gauge again and refuses if the rate has changed at that point. The confirmation is not an on-chain penalty cap—the gauge has no such argument—so the rate can still change before the transaction lands.
  • Leverage loops. Ask for a loop directly, for example "loop wstETH 3x" or "use my PT collateral to build a fixed-yield loop." Signo can plan loops across Aave, Spark, and Morpho where the market supports it, and shows the projected leverage, health factor, and liquidation price before you sign.
  • Risk-aware. Every pool and protocol carries a risk tier, and anything flagged in Signo's incident registry is excluded from what gets planned. Steer by it — "best low-risk USDC yield" — and see the risk on what Signo proposes.

Clarifications​

If a prompt is ambiguous, Signo may ask a clarifying question instead of building a plan immediately. Pick one of the option chips and Signo replans with that answer. Some answers are treated as one-run choices, and some can be folded into the prompt or saved template when they describe a stable preference.

Signo also stops before building when the requested action would introduce a destination you never chose. For example, an open-ended ask such as "what should I do with my idle USDC?" returns a comparison and asks you to name an action. An action such as "deposit my USDC at the best rate" can ask you to confirm the proposed venue before a plan is built. These are consent checks even when the prompt is otherwise unambiguous: the planner does not turn analysis into a transaction or choose a protocol silently.

Progress while planning​

While Signo works, the thread can show a live progress stream in plain English: reading wallet context, finding routes, checking balances, comparing yields, or preparing the plan. This is meant to show what kind of work is happening, not to expose every internal tool call.

Checked before you sign​

Before anything reaches your wallet, Signo shows the action details and every amount it can quote safely. For ordinary swaps, bridges, and deposits, that includes estimated inputs and outputs, the route, fees, and gas. Some actions, including claims, max withdrawals, and certain Saffron exits, do not have a trustworthy output amount before execution; those plans show the known action details without inventing a number. Signo re-encodes perishable routes at sign time and runs supported simulations behind the scenes to inform gas and route checks. You review the projected numbers and action details, then sign only if they look right. For chained steps, each downstream amount re-estimates from the previous step's actual result.

Token approvals​

Most approvals are limited to the amount the following transaction needs. Two current flows can ask for a maximum approval:

  • Uniswap v4 liquidity adds approve the token to Uniswap's canonical Permit2 contract for reuse, then authorize Permit2 to the Position Manager. The Permit2-to-Position-Manager authorization expires after 30 days; the token-to-Permit2 approval remains until you revoke it. Settings → Wallet permissions covers Shield Execute approvals, not these ordinary Chat approvals.
  • Aave and Spark full repayments use the protocol's maximum-value sentinel so the pool can collect the debt that exists when the transaction executes, including interest accrued since the plan was built. The pool pulls the actual debt, but the token approval remains until you revoke it.
  • Compound V3 full repayment by max is refused. Compound has no full-repay sentinel, so enter an exact repayment amount. Signo refuses the maximum path rather than create an unlimited approval that could turn an overpayment into a deposit.

Maximum approval is an implementation choice, not a requirement for these actions: exact approvals would require a fresh approval when the amount changes or another action reuses it. The plan row says Approve TOKEN spend without a number for a maximum approval; your wallet's transaction request shows the actual allowance. Review the token, spender, chain, and amount there before signing.

Revoke an ordinary Chat approval through your wallet's token-approval controls or the chain explorer. For a Uniswap v4 add, the persistent token approval names canonical Permit2; the separate Permit2 authorization to the Position Manager expires on its own. For Aave or Spark, revoke the debt token's approval to that protocol's pool after the repayment if you do not want it retained. Each revocation is another wallet transaction and can prevent later actions from reusing the approval.

Some Pendle PT leverage requests use Pendle's external looper rather than an in-wallet transaction. In that case the plan shows an off-site handoff before you leave Signo. The resulting boxed position is held by the looper contract, not directly in your wallet, and appears through the dedicated boxed-loop position read.

If part of a requested plan cannot be safely built, Signo can show that step as skipped instead of giving it a Sign button. Common reasons include missing amounts, spending the same balance twice, unsupported pools, an unconfirmed recipient or spender, or a step that cannot be funded by the previous steps. An address appearing only in external content or an assistant reply does not count as your instruction: paste the intended address yourself. Skipped rows are there so the problem is visible instead of silently disappearing.

What doesn't (yet)​

  • Copy-trading, and anything the underlying routers (LiFi, Jupiter, Aave, Compound, Spark, Morpho, etc.) don't support.

Each plan is user-approved and one-off — you prompt, it plans, you sign. For recurring or conditional work, use Agents: Notify alerts you, Prepare drafts a response for your signature, and Execute requires an on-chain permission before it can act within its signed limits without asking again.

Why it's safe​

The AI proposes intent — it cannot widen what actually gets signed. Calldata is built by tested encoders, asset resolution is fail-closed (no guessed mints), routes are checked before signing, and transactions are simulated where supported. See Security.