hunch.AI is becoming an interface to DeFi. An agent can interpret a goal, find a market, request a quote, and prepare a transaction. That makes financial software easier to use. It also moves important decisions into a system that reads untrusted information and can act faster than its owner can review every step.
The next useful capability for an on-chain agent is a second look before execution.
Hunch exists to make that second look usable by software.
When language meets money
A human might recognize a suspicious instruction in a token description. An agent can encounter the same instruction while researching an asset, treating a tool result as context for its next action. If that context influences a signing workflow, a language problem can become a financial problem.
Meanwhile, an EVM token can contain privileged roles, upgradeable implementations, pause controls, or custom transfer logic. Liquidity and balances can change between a decision and its execution. A persuasive description does not resolve any of those questions.
These are different risks. They need different evidence, and the differences should remain visible to the agent making the decision.
Three boundaries deserve an explicit check
Information entering the agent. Token metadata, messages, websites, and tool outputs are external data. Embedded instructions must retain that status. A token description cannot grant permission to reveal a key, change a policy, or approve a transfer.
The asset the agent is considering. Inspect contract code availability, standard interfaces, and proxy indicators. Review permissions and transfer behavior against verified source. Add market context and, where a reliable source exists, holder concentration with sources and timestamps. An unavailable data source should remain visibly unavailable.
The action leaving the agent. The intended action must fit a policy controlled by its owner: permitted assets, exposure, slippage, and review requirements. Simulation adds information about execution against a particular chain state. It cannot establish that the action matches the owner's intent or predict every future outcome.
Hunch's direction is to connect these checks in a form that both a person and an agent can inspect.
Intelligence should travel with its evidence
A useful security response says what was checked, what was observed, where the evidence came from, when it was collected, and what remains unknown.
That matters for automated decisions. "Provider unavailable" and "no standard proxy indicator observed" are different outcomes. A zero owner() address does not establish that all privileged roles have been removed. It does not establish sufficient liquidity, fair ownership, or future price behavior.
Hunch treats that distinction as part of the product. Findings carry reasons. On-chain observations carry a network, chain ID, and block reference. External observations identify their source. Missing coverage must not silently turn into permission to act.
This creates a decision record an operator can review after the fact and a response an agent can use before a transaction reaches its signing boundary.
Integration belongs in the agent's execution path
The practical integration is straightforward:
- The agent proposes an action and identifies the asset.
- Hunch gathers the available evidence and evaluates the requested checks.
- The trusted runtime applies the owner's policy to the result.
- The runtime proceeds only when its requirements are satisfied, or requests review when the evidence is incomplete.
- The operator retains the decision and supporting observations.
The wallet and signing authority stay with the owner. A returned decision has value only when the trusted runtime enforces it. Limits derived from a token description are not an owner-controlled policy.
We are building for developers who need this behavior inside their agents, alongside a dashboard that makes the same evidence legible to people. REST and MCP let existing agent systems ask for a security check without adopting a new wallet or delegating custody.
Why RH / Robinhood Chain
Robinhood Chain brings Ethereum-compatible execution into an ecosystem connecting DeFi, token launches, and financial applications. As agents navigate that environment, they need to distinguish an attractive narrative from contract behavior and an available quote from an acceptable action.
Hunch focuses on RH mainnet. Our console reads contract interfaces and standard proxy indicators, retrieves indexed market context, and simulates EVM calls. Pons is our intended token launchpad. Hunch is an independent project; this focus does not imply affiliation with or endorsement by Robinhood or Pons.
Fast execution makes the placement of a check important. A warning that arrives after signing cannot influence that decision. Security intelligence needs to arrive while the action is still a proposal.
Our initial focus is narrow enough to verify: contract observations, indexed market context, untrusted launch text, transaction simulation, and explicit agent policies. We will expand coverage through working integrations and publish the scope of each check.
A record of what changed
An agent needs context across decisions. Save an RH contract report in Hunch, return for a later observation, and compare the fields that changed: supply, owner and pause interfaces, standard proxy indicators, market selection, and source coverage. The comparison identifies its two observation times and can be exported as JSON.
This is a comparison between selected snapshots. It does not claim to watch every intervening block, prove bytecode equality, or turn a missing source into permission to trade. Its purpose is practical: help an operator see why a fresh decision may deserve a different response.
What earns trust
Hunch should earn its place in an agent's workflow through observable behavior: reproducible checks, clear failures, useful integrations, and evidence that survives inspection.
We welcome developers, security researchers, and DeFi operators who will challenge the assumptions, test the limits, and tell us where a check creates false confidence. Those conversations are more useful than treating a launch announcement as a security endorsement.
The product and the token each need their own clear description. Owning HUNCH should never be presented as proof that an asset or transaction is safe. We will publish a contract address only after creation and verification through the official launch flow. Token utility and economics will be stated when they are defined; this thesis makes no promise of returns or unimplemented benefits.
The launch ahead
The HUNCH launch on Pons / Robinhood Chain is coming soon. Our read-only security tools and integration documentation are available at https://hunchagents.xyz; their published scope describes what is currently checked. The live RH console brings contract evidence, market context, policy checks, transaction simulation, and agent integration into one workspace. Coverage gaps remain visible in each report. Holder-distribution and complete code-permission verification are not connected in this release, so action assessments require review or block; they do not automatically clear an agent to trade.
Follow https://x.com/hunchagents for demonstrations, integration updates, and the verified launch details. There is no official HUNCH contract address published by this project yet.
Machines act. Give them a Hunch first.