Enterprise chaos
Enterprises rarely feel AI cost the way a solo builder feels an OpenAI invoice. Spend hides in commits, shared keys, and quarterly FinOps. What they cannot ignore is chaos: shadow tools, the same prompts billed twice, unsafe browse, opaque multi-vendor bills, unattended agent sessions nobody reviewed, and teams locked to one IDE.
Metrecept is the black box for that chaos, at organizational scale — one OpenAI-compatible pipe, org SSO, compliance policy, and a corporate ledger where every line is a signed, independently verifiable receipt, not just a billing rollup. Coding agents (Cursor, Claude Code, VS Code, and friends) plug in over MCP when you want them; SDKs and the Shell work without them.
The chaos map
| What's going wrong | What it looks like | What Metrecept does |
|---|---|---|
| Repeat inference | Agents retry the same prompt; bills and rate limits spike | Exact-match Redis replay + receipted meters on HIT/MISS |
| Unaccountable agent sessions | Nobody reviewed what an unattended run called or cost | Signed session receipt aggregating every event |
| Shadow AI | Personal keys, unmanaged chat tools, no allowlist | Org tenancy, SSO, enforced ingress, model policy |
| Unsafe browse | Scrapers stuffing junk (or PII) into prompts | Purpose-bound, robots-aware public-web ingest, receipted |
| Opaque vendors | Three lab invoices, no chargeback story | Cost-center ledger + monthly FinOps statement |
| Client lock-in | "We can only work inside one IDE" | Agent Shell + any base_url client |
| Procurement fear | No admin surface, no audit, weak DPA story | Org console, audit log, enterprise pack |
Product surfaces
| Surface | Role |
|---|---|
https://api.withohm.dev/v1 | OpenAI-compatible pipe (BYOK or managed pool) |
| Org console | Members, cost centers, policy, ledger / month CSV |
| Session demo | Signed, verifiable receipts for a real agent session |
| Integrations | Cursor, Claude Code, VS Code, Windsurf, Zed, and the pipe stack |
| Enterprise apply | Dedicated-pool SKU — contact us |
How a platform lead proves it
SSO in → mint a service key → attribute spend to two cost centers → export a month → show Legal a denied fetch or a signed receipt for a specific incident. All of that lives in the org console and the pipe — preferred clients stay on the integrations board.
Tag agents with X-Ohm-Path (docs-bot, ci-prompts, …). Org spend caps soft- or
hard-stop MISS flood per cost center; HITs still serve. FinOps stays
estimate_only — provider invoice import is not shipped.
Internal paths (where exact-match density is high)
| Path | Why hits accumulate |
|---|---|
| Docs Q&A bots | Same retrieval + answer template, many times a day |
| Support triage | Short classified prompts over recurring ticket classes |
| Branch / preview automation | Mechanical agent loops with stable tool prompts |
| CI prompt suites | Byte-identical checks on every PR |
Point those agents at the Metrecept base_url (or attach MCP), keep prompts
byte-identical, send a stable X-Ohm-Path, and the ledger records HIT vs MISS
by cost center and path — each one receipted, not just tallied.