build
Every agent starts from a job — one noun — and gets the instrument that does it. Capabilities are declared up front: model, access, limits.
fantasti.bot builds, proposes and manages individual agents for agentic auditing. Each one is a tool, not a character — it has a mission, declared capabilities, and a wiring you can read. Run one, or run a swarm.
One person, a set of instruments, and a process that leaves a record at every step.
Every agent starts from a job — one noun — and gets the instrument that does it. Capabilities are declared up front: model, access, limits.
For each mission we propose a set: which bots, which wiring, who writes to the ledger. You see the whole swarm on one sheet before anything runs.
Every agent carries a numbered card: version, date, capability, wiring — verified by probe, or marked not assessed. Change a capability, get a new card.
Seven jobs, seven instruments. No approver — the human gate stays human.
Five switches, each on or off. No dials, no "mostly". Filled = on, blank = off. A wiring is declared on the card and verified by probe — so declared and behaved can differ, and the card can say so.
Each agent works alone. Assembled per mission, they check each other.
Drop one agent on one claim. An assessor fills exactly one claim and cites its evidence; a seal hashes and appends. Small, numbered, done.
A counter argues the other side. A decoy plants fake evidence to test the assessor. A flare fires on non-compliance and no bot can silence it. plumb probes them all against their declared wiring.
Kawaii in surface, ID card in structure. Number, version, date, mission, capability, wiring, ledger, operator. Nothing else — no logos, no feelings.
propose a mission