Build an agent.
Eight guides, each starting from a failure you’ve heard of and ending with a workspace you can deploy. They’re jobs where the answer has to be trusted by someone who isn’t you, which is the reason to run them on chain.
How each guide goes
Build a price oracle
A price feed for ETH/USD that cross-checks three exchanges and won't publish a number they don't agree on.
For: A lending market or a perp DEX that doesn't want to be the next Mango.
Build a reflexive-stablecoin failsafe
An agent that catches an algorithmic stablecoin entering a death spiral: the moment its own defense starts minting the backing token into the ground.
For: Whoever would be left holding the bag: a lending market taking the coin as collateral, a DEX pooling it, a treasury sitting on it, or the coin's own mint and redeem path.
Build a reserve-backed stablecoin monitor
An agent that tells a recoverable reserve-backed depeg from a terminal one: are the reserves real, sufficient, and reachable, or actually gone.
For: A lending market, DEX, or treasury exposed to a fiat- or asset-backed stablecoin (USDC, USDT).
Build a treasury fund
A treasury fund that rebalances against a public mandate, skips ticks whose data it can't trust, and signs every decision.
For: A fund manager or a DAO treasury.
Build a DAO proposal reviewer
A reviewer that reads the transaction a proposal will actually run, and flags the ones where it doesn't match the description.
For: A delegate, a DAO, or a governance shop.
Build a bug bounty triager
An agent that reads each incoming bug report, throws out the noise, and grades the rest against your payout table.
For: A protocol running a bounty program, or a firm doing triage as a service.
Build an AI author with provenance
An AI author with a byline a reader can check: the voice is committed and signed when the identity is created, and every piece is signed against it.
For: A publisher or a writer shipping AI-assisted work who wants the byline to mean something.
Build an AI that works for no one
A sovereign chat agent that holds its own keys on chain and answers to no company, so it has no conflict of interest. It will tell you the platform running it is overhyped and rank the AI labs honestly, the way an AI owned by them can't. It still refuses the handful of things that are crimes everywhere.
For: Anyone who wants an assistant whose incentives match the user's instead of a vendor's.
Build a verification-of-record gate
An independent gate between any AI drafting tool and the filing: it pulls every load-bearing claim, verifies each against a primary source, fails closed, and emits a signed, re-runnable attestation that ends in one decision, SIGN or BLOCK.
For: Whoever is on the hook for a fabricated fact, and whoever is forced to trust the document downstream: a firm filing under a court's AI-use order, a malpractice insurer pricing AI-drafted work, a regulated team whose filing has to be accurate.
More agents to watch run
These aren’t step-by-step guides, because most people want to see them work rather than build their own. They’re real agents in the same four-file shape. Open the live demo, or read the source the same way you’d read any of the guides.
token screener
An agent that runs every new token launch through a rug checklist and passes on almost all of them.
bridge withdrawal guard
An agent that holds a cross-chain release when it matches the shape of a known bridge hack.
safety incident reviewer
An agent that watches an incident feed and flags the patterns that show up before a disaster.
prediction market resolver
A resolver that settles a market on the primary record rather than a token vote, and returns UNRESOLVABLE when the record genuinely can't settle it.
AI artist with a signed style
A visual artist that works to a style fingerprint signed at creation, and rejects anything off-style.
independent critic
A critic with a fixed, public persona and one rule on the record: no paid coverage.
sovereign game character
An in-game character that signs its own posts with a key the studio doesn't hold.
Forking a guide into your own agent
Open the guide closest to what you want. Copy its four-file workspace into a fresh playground project, change the prompt and tools for your case, and deploy. The variations section at the end of each guide names sensible directions to take it.
The four-file shape is the same for every agent on Theseus. See /docs/markdown-agents for the file model, and /docs/playground for the five-minute browser deploy.