Governance Ledger
Every model call our own AI infrastructure makes is logged, costed, budget-capped, and attributed — automatically, as a byproduct of how it routes work. This is a real excerpt from that ledger.
Spend vs. cap
$0.000958
of $30.00 · 30-day rolling window
Calls logged
17
unsampled — every call, not a spot check
Backends in rotation
3
local free-tier → Google → OpenRouter, by policy
Paid vs. free-tier
2 / 17
88% of calls settled at $0 before touching billed credit
Per-call ledger
What was requested, what actually served it, what it cost, and which governance function it evidences. Nothing here was written for this page.
| Timestamp | Requested → served | Tokens in/out | Cost | Pool | Quality | Evidences |
|---|---|---|---|---|---|---|
| Jul 17 16:54:27 | gemini-flash→gemini-3-flash-preview | 29 / 6 | $0.000002 | free-tier | — | MAP |
| Jul 17 16:54:37 | glm-5.2→z-ai/glm-5.2 | 43 / 150 | $0.000490 | OpenRouter | — | MAPMANAGE |
| Jul 17 17:20:09 | glm-flash→qwen/qwen3-8b | 14 / 30 | $0.000000 | free-tier | — | MAP |
| Jul 17 17:20:46 | glm-flash→qwen/qwen3-8b | 14 / 400 | $0.000000 | free-tier | — | MAP |
| Jul 17 17:20:49 | gemini-flash→gemini-3-flash-preview | 8 / 20 | $0.000003 | free-tier | — | MAP |
| Jul 17 17:21:18 | glm-flash→qwen/qwen3-8b | 14 / 400 | $0.000000 | free-tier | — | MAP |
| Jul 17 17:21:20 | gemini-flash→gemini-3-flash-preview | 8 / 20 | $0.000003 | free-tier | — | MAP |
| Jul 17 17:21:45 | glm-flash→qwen/qwen3-8b | 18 / 13 | $0.000000 | free-tier | — | MAP |
| Jul 17 17:21:48 | gemini-flash→gemini-3-flash-preview | 8 / 1 | $0.000000 | free-tier | — | MAP |
| Jul 17 17:21:52 | glm-5.2→z-ai/glm-5.2 | 19 / 79 | $0.000255 | OpenRouter | — | MAPMANAGE |
| Jul 17 17:54:32 | glm-flash→qwen/qwen3-8b | 17 / 12 | $0.000000 | free-tier | — | MAP |
| Jul 22 22:39:58 | glm-flash→qwen/qwen3-8b | 19 / 6 | $0.000000 | free-tier | — | MAP |
| Jul 24 12:57:27 | glm-flash→qwen/qwen3-8b | 21 / 7 | $0.000000 | free-tier | — | MAP |
| Jul 25 11:15:59 | glm-flash→qwen/qwen3-8b | 17 / 12 | $0.000000 | free-tier | — | MAP |
| Jul 25 14:07:30 | glm-flash→qwen/qwen3-8b | 18 / 7 | $0.000000 | free-tier | 0.90 | MAPMEASURE |
| Jul 25 15:14:39 | glm-flash→qwen/qwen3-8b | 19 / 8 | $0.000000 | free-tier | — | MAP |
| Jul 25 15:34:07 | glm-flash→qwen/qwen3-8b | 19 / 6 | $0.000000 | free-tier | — | MAP |
Live gateway snapshot
Checked directly against the gateway's own status tool on Jul 27, 2026 — a separate, fresher check than the ledger rows above, not a live-refreshing counter on this page.
Monthly budget cap
$30.00
hard ceiling, shared across every machine running this infrastructure
Spent this window
$0.0007
0.002% of cap — window opened Jul 17, 2026
Per-request ceiling
$1.00
reserved and checked before any single call goes out
Backend health
3 / 3
local T0, Google, and OpenRouter all reachable at last check
What this satisfies
Mapped to the NIST AI RMF's four functions — the framework this ledger was built against, not retrofitted to.
GOVERN
A spend policy — $30 per 30-day window, shared across every machine running this infrastructure — set once and enforced automatically on every call.
policy: budget.json
MAP
Every call records what was requested and what actually served it. No AI action happens without a named model and a named route.
field: model → served_model
MEASURE
Token volume and cost are recorded on every call; response quality is scored and attached where evaluated.
field: in / out / cost / quality
MANAGE
Spend against paid credit is checked and reserved before the call goes out — the system can't overspend the cap, because it never sends a request that would.
function: budget reserve-and-charge
This is the pattern we build into client systems: every AI action logged, budgeted, and attributable — applied to a governed write-path into your ERP, or packaged as evidence for an insurance renewal, instead of a general-purpose AI gateway.
Want the full picture for your environment?
A discovery call gets you a scoped assessment from the team that builds these migrations — not a form, a conversation.
Book a discovery call