Durable workflows
Persist actions, checkpoints, step state, and resumable progress across restarts. No lost context when a task pauses.
Durable, resumable workflows for agents that need to call APIs, request approval, move value, and prove exactly what happened.
FlowMint separates orchestration from signing, so automation can move quickly without becoming an unchecked payment rail.
Persist actions, checkpoints, step state, and resumable progress across restarts. No lost context when a task pauses.
Pause risky steps at waiting_approval. A human can inspect, approve, reject, and leave a note.
Turn an HTTP 402 response into a payment intent, policy check, approval state, and confirmable receipt.
Allow, warn, or block by amount, recipient, network, and cumulative limits—before a wallet signs anything.
Every event is hash-chained with a root hash ready for NEAR anchor preparation and independent audit.
Untrusted instructions cannot override payment policy. Blocked attempts are recorded with reasons and funds sent: 0.
Choose a scenario. The terminal is an honest, offline simulation of the real FlowMint state transitions—safe for a live demo.
Use FlowMint wherever an agent needs durable progress, controlled side effects, and a defensible audit trail.
Let an agent buy a metered API response through x402 without giving it unrestricted wallet authority.
Run multi-step research, pause for editorial approval, resume later, and preserve the complete provenance chain.
Require recipient allowlists, spending caps, and human approval before an automated purchase can proceed.
Coordinate actions that survive restarts while keeping every external side effect reviewable.
Test whether prompt-injected instructions can cause an unauthorized transfer—and show the block as evidence.
Prepare quote and settle requests for a facilitator while the wallet or managed signing runtime stays outside FlowMint.
Orchestration remains durable; signing remains non-custodial and policy-aware.
Define steps, tags, policy, and approval requirements.
Evaluate amount, network, recipient, and context.
Pause for a human or deterministic policy decision.
Append receipts, compute root hash, prepare anchor.
Explore the offline demo, inspect the source, and bring policy-controlled agent workflows to your next application.