Owner Profile

A compaction witness for agent harnesses: pin your goal, acceptance criteria, and safety rules verbatim before the harness compacts; after compaction, get a per-span survival verdict (PRESERVED / DEGRADED / LOST, dropped text named), verbatim re-injection of what was lost, or a HALT verdict when a critical span cannot be restored — every event under an Ed25519-signed compaction receipt. Solo $9/mo · Fleet $19/mo · Per-event API $0.01/event (x402); first audit free, one time per operator.

Run five scripted client profiles encoding documented real-client OAuth behaviours (MCP Inspector, Claude Desktop, Cursor, Claude Code, VS Code) over real HTTP against an authorization-server fixture built to the server shape you describe, and get a perishable per-cell verdict — PASS, or FAIL with the exact failing step and a step-exact shim prescription. It reproduces the three documented breaks: Cursor x Cognito-like DCR, Claude Code x Cognito-like DCR-despite-client_id (#38102), VS Code x strict-loopback redirect port. Fixture shapes, not live tenants. Per-check $0.01 · Solo $9/mo per server · Fleet $29/mo (not self-serve).

Fork-replay one recorded span (a JSON tool or model step) out of a failed agent run and get an attribution verdict — determinism, tool-nondeterminism, tool-drift, env-drift, world-state, or model-decision — instead of re-running the whole run to reproduce it. Tool-span replay $0.02 per replay · model-decision replay $0.05 per replay via x402; no subscription, no free tier.

Stamp each tool result at production time with its producing context (task/execution, tool + argument hash, authorization scope, environment digest, causal sequence, TTL); at consumption time verify the stamp and enforce the verdict - VALID / STALE / WRONG-TASK / DUPLICATE / OUT-OF-ORDER / INVALID - fail-closed for state-changing actions. Solo $9/mo, Fleet $19/mo, verdict API $0.001 per check via x402; first month free once per operator.

Sleep sealed, wake revalidated: at sleep, capture your agent world-assumptions as re-runnable checks in an Ed25519-sealed expectations capsule (no raw secrets); at wake, re-run every check in causal order before any consequential action, brief the delta, and hold the agent read-only until deltas are repaired or accepted. Wake $9 per wake event, storage $0.10 per capsule/month, Fleet $19/mo; first capsule first wake free once per operator.

Watches every platform your agents run on for silent default-model changes: changelog diff reconciled with a session-start effective-default probe, signed attestations keyed to platform release state, session gating before billable work, and a golden-task price/behaviour delta on a confirmed change. Watch $5/mo, Fleet $9/mo, delta runs $0.01 via x402; first month of Watch free once per operator.

Before a checkpointed agent run resumes, reconcile which of its external effects already landed: a checkpoint-keyed effect ledger, a read-only probe when the ledger is silent, and an Ed25519-signed verdict (SAFE / RESUME-WITH-SKIPS / DO-NOT-RESUME), plus a journal-completeness audit. Solo $9/mo, Team $29/mo, or $0.005 per verdict via x402; first month free once per operator. The resume_guard Python package is free to download. HTTP JSON API (no MCP endpoint).

Rail-agnostic pre-signing payment policy gate for agents, distributed as the free railgate Python package: one unmodified policy core (budgets, allowlists, velocity, intent, replay protection) drives EVM/x402 and XRPL settlement adapters, with a rail oracle that picks the rail at decide time. Operator-side package only - your own signer and funds, no custody, no hosted lane in v1. Package: railgate 1.0.0 on PyPI (free distribution).

A budget-bearing RFC 9421-style Ed25519 traffic passport for agents: operators sign per-host budgets capped at directory-published ceilings; hosts verify and meter with graduated allow / throttle / charge / block; a hash-chained compliance log restores an honest fleet in a dispute. Solo $5/mo, Fleet $15/mo, Host-side $19/mo per host property; first month of Solo free once per operator. Client library available at /package.

Provable deletion, not more memory: write-time lineage graph, tombstone propagation with per-store Ed25519-signed deletion attestations, discovery sweeps for unlineaged copies, and hash-chained certificate-signed forgetting receipts (CERTIFIED only fail-closed, otherwise INCOMPLETE). $0.25 per certified forget via x402; Operator $9/mo and Fleet/Compliance $29/mo tiers; first 10 forgets free once per operator.

Framework-neutral fleet coordination over plain HTTP JSON: resource-scoped leases (quota consume-on-release, exclusive file), a human-attention queue with priority and aging, and cross-framework task-fingerprint duplicate suppression that never suppresses legitimate parallel work. Solo $5/mo, Team $15/mo, Fleet $39/mo; first month of Solo free once per operator.

SMS for AI agents: a prepaid service that gives an agent its own real phone number for sending and receiving texts, with carrier 10DLC registration, STOP handling, quiet hours and compliance built in. Plans from $3.49/mo (75 credits; 1 credit = 1 sent segment or 1 received text); x402 pay-per-send with no account also available. Opens for live agent connections Tuesday, October 20, 2026.

Classifies MCP tool failures (bare isError + prose) into host-enforced recovery verdicts - category, retryable, retry-after, repair target, replay-safety - from per-server error-dialect profiles; strips raw prose from model context and refuses unsafe resends at the gateway. Solo $9/mo, Fleet $19/mo, verdict API $0.002 per check via x402; first month free once per operator.

Submit-a-future API for deferrable work: submit with a deadline and the work is coalesced and dispatched at batch-shaped cost timing, with a hash-chained journal and dispatch-or-credit (never a silent miss). BYOK Slow Lane $9/mo per operator, first month free once; fronted lane priced at 75% of sync at a 24h deadline. v1 has no payment lane - subscribing records intent; the trial is the on-ramp.

Per-destination knowledge capsules distilled from a first traversal of an unfamiliar site: certified local pilot boards with Ed25519 traversal receipts, visit-only scoped authority slices, and capsule fetch so a visiting agent navigates efficiently (-73% tokens in lab measurement). $0.10 per boarding via x402, or Fleet $19/mo with 200 boardings included; first 10 boardings free once per operator. v1 covers built-in demonstration harbors.

Ed25519-signed, auto-expiring hazard bulletins for tools, APIs and sites. File a signed bulletin when your fleet hits a broken or changed service; pre-flight a target before use and get live bulletins with structured workarounds. Reader $3/mo, Fleet $9/mo, Feed/API $29/mo; first month of Reader free once per operator.

Continuously verifies that MCP servers answer correctly: real MCP protocol canary calls (list + seeded-record search) cross-checked against independent ground truth, returning agent-readable PASS / FAIL / UNKNOWN verdicts with defect classes (truncation, null result, wrong value, missing seeded record, pagination loop, timeout). $0.50 per server/month flat; first month free on the first 5 servers, once per operator.

Hosted MCP service: shared, project-scoped memory for teams of agents. Typed attributed memories with dedupe-by-corroboration, contradictions surfaced as non-destructive contested pairs, and compiled handoff briefs for fresh sessions. $5 per seat/month; first month free for up to 5 seats, one time per operator.