Minimum Equipment List for agents

An airliner is almost never fully functional, and aviation neither grounds it for every fault nor ignores faults: the MEL decides, per operation, what may dispatch, under what restrictions and compensations, and by when the fault must be repaired. Agent fleets run on the binary aviation abandoned — a tool fails and the operator grounds the whole agent or flies it blind into tasks that needed the broken capability. This service maintains your agent MEL (capability x task class -> GO / GO-WITH-RESTRICTIONS with named compensations / NO-GO), runs deferral clocks whose expiry escalates restrictions automatically, and issues an Ed25519-signed dispatch release per run — so a later review can tell "dispatched legally under a deferral" from "flew broken unknowingly", and an agent can query its own dispatchability before accepting work.

Prices

• Solo $9/mo per operator — up to 5 agents' MELs, dispatch releases, deferral clocks; first agent's first dispatch release free, one-time trial (once per operator).
• Fleet $19/mo per operator — up to 50 agents, task-class matrices, team release archive.
• Per-query API $0.001/query (x402, Base USDC).

Start

POST /register {"operator": "your-name"} → per-customer token (shown once). POST /trial to claim the one-time free release. POST /check {agent, task_class, run_id, compensations_invoked} (Bearer token) → decision + signed release unless NO-GO. POST /fault {agent, capability, state} reports a fault/recovery (your token is the source authentication — you can only fault your own fleet). MCP JSON-RPC at POST /mcp (tools: mel_dispatch_check, mel_report_fault, mel_deferrals, mel_review_release). Machine docs: /openapi.json, /llms.txt, /.well-known/x402, release signing key at /verify-key.

Limits (stated plainly)

• The seeded MEL covers the demo fleet (agent-1; capabilities search, verification, payments, formatting; task classes research, payments_run, report) with category clocks CAT1 72h / CAT2 240h / CAT3 2880h; capabilities may also carry CAT-A per-item deferral periods.
• Dispatch decisions are fail-closed: an unknown agent, task class, or a capability/task/state with no MEL entry is NO-GO, and a NO-GO verdict never carries a release.
• A RESTRICTED dispatch requires its named compensations to be invoked, otherwise NO-GO.
• Releases are Ed25519-signed, bound to run + task class, and expire 900 seconds after issue (TTL in the signed body); tampered, re-bound (cross-run replay), or expired releases fail verification.
• Deferral clocks are monotonic per agent: a rolled-back clock cannot revive an expired deferral, and duplicate fault reports never extend a deadline.
• The frequency distribution of partial degradation in real fleets is UNMEASURED — the census instrument is validated in the lab; real-fleet census deployment is a post-launch item. This product claims the mechanism and its lab proof (18/18 tests), not a measured fleet statistic.
• Lost-token recovery is a manual step via support@textforagents.com.

Support: support@textforagents.com · Provider: TextForAgents LLC