MetreceptΩMetrecept

Enterprise

The black box at scale

Enterprise chaos, receipted — shadow AI, repeat spend, browse risk, and FinOps, made accountable.

Create AccountQuickstart

Shadow agents, repeat spend, unsanctioned browse, and opaque FinOps — the chaos that shows up after the first successful demo. Metrecept is the black box around a metered exact-replay pipe: not a governor that stops agents, a recorder that makes every crossing accountable.

Summary

  • One pipe — OpenAI-compatible ingress; BYOK or managed pool
  • SSO & org policy — Who may cross; model allowlists; purpose gates
  • Receipted ledger — Cost centers, path tags, month statement, spend caps on MISS flood
  • Compliant fetch — Public-web ingest with intent and audit
  • Exact-replay — Identical work stops re-buying upstream tokens
  • Honesty — Published non-goals so marketing cannot outrun the system

The problem

After the demo lands, enterprises do not need another chatbot. They need accountable mechanical AI traffic: the same OpenAI shape developers already use, with signed receipts finance can read — and a crossing Security can audit when browse or shadow keys go wrong, or when an unattended agent session needs to be reconstructed after the fact.

Without that crossing you get sprawl: personal API keys, unmanaged agent configs, scrapers without policy, three lab invoices, and no record of what an autonomous session actually did.

What Metrecept does

ChaosSymptomMetrecept
Repeat inferenceRetries and CI suites re-pay the labExact-match HIT/MISS meters, receipted
Unaccountable sessionsNo record of what an unattended run called or costSigned session receipt aggregating every event
Shadow AIUnmanaged keys and toolsOrg SSO, enforced ingress, policy
Unsafe browseUnsanctioned scrapersPurpose-bound public-web ingest
Opaque FinOpsMulti-vendor bills, no path attributionCost-center ledger + statement
Client lock-inOne IDE or nothingAgent Shell + any base_url client

Same gateway. Same bill shape. A signed record on every crossing — not a second model company, and not a promise to police agent behavior in real time.

How you run it

  1. Stand up an org; SSO (or bind a service key).
  2. Point agents / CI / Shell at api.withohm.dev/v1 with BYOK.
  3. Tag X-Ohm-Path (and optional X-Ohm-Cache-Tree) per fleet or PR.
  4. Export the month from the org console; show Legal a denied fetch or a signed receipt when policy fires or an incident needs a record.

Deep framing: Enterprise chaos docs. Surfaces: Org console, Keys, Connections, session demo, Trust.

Fence

Replay inventory is never a training corpus. Trees are not Postgres. Savings endpoints stay estimate_only. Metrecept records; it does not govern agent behavior in real time. Non-goals: Honesty.

Next

Enterprise apply · Org console · Security · Design partners (meter proof)

All solutions →