Tarmac / Product
Phase 1 ships Month 7 · MVP behind the design-partner cohort

One ledger. Two views. Every dollar accounted for, before it's spent.

Tarmac is a single control plane with two reconciled surfaces: an operations dashboard for the platform team, and a finance-grade console for the CFO. Both read from the same atomic authorization ledger — there is no out-of-sync report and no end-of-month reconciliation.

// 01 · Architecture

Tarmac sits between your agents and the compute they consume. Proxy in path. Ledger as source of truth.

Every agent request flows through the Enforcement Proxy. Every authorization is decided by the Budget Engine against the live ledger. Every spend lands on the provider — but only after an atomic hold is placed.

// In your network

Your agent fleet

Coding, research, monitoring, data-pipeline agents — built on LangChain, CrewAI, raw SDKs, or your own orchestrator. No code change beyond the base URL.

support-triage · P0
revenue-ops · P0
data-pipeline-eu · P1
+ 14 more
// Tarmac control plane

Enforcement Proxy + Budget Engine

Authorization holds are placed atomically against your live budget. Decisions are made in single-digit milliseconds. The ledger is the system of record.

hold $4.10 → ledger
decide tier · cap fit
admit · forward request
settle $1.20 · release $2.90
// Your provider accounts

Compute providers

Inference-API lane (Anthropic, OpenAI, Google, Bedrock, Azure OpenAI) ships first. GPU-rental lane follows as an adapter — same price signal, same enforcement.

Anthropic · Claude family
OpenAI · GPT family
Google · Gemini family
Bedrock · multi-vendor
Proxy-first. Short-lived scoped tokens · no long-lived customer credentials at rest. · Fail-closed. If the engine is unhealthy, the proxy refuses to admit. · Concurrency-invariant. CI release gate — no two runs consume the same dollar.
// 02 · Surface 01

Agent Operations Dashboard. The live control room for platform engineering.

// Surface 01 of 02
For platform & AI-infra leaders

Watch every authorization. Rebalance before a cap is breached.

The Operations Dashboard is the platform leader's home screen. Admits, holds, throttles and tier moves are visible the moment they happen — not surfaced after a billing window closes. Drilldowns go all the way down to the single authorization decision and the policy rule that produced it.

01

Fleet view

Every agent, its priority tier, its active model tier, hold rate and live burn — sortable, filterable, deep-linkable.

02

Authorization timeline

Stream of admits, throttles, and holds with the rule that fired, the budget remaining at decision time, and the resulting receipt.

03

Priority rebalance

When the monthly cap tightens, Tarmac sheds P3 work first, then P2 — automatically. P0 keeps its full model tier.

04

Per-agent rules

Soft alerts, hard floors, route-around rules, per-tool quotas. Edit live; rules take effect on the next authorization.

05

Drift & anomaly

Detects an agent whose mean cost-per-run shifted > 2σ. Catches the runaway loop or new tool spend before it becomes a hold spike.

06

Webhooks & integrations

Slack, PagerDuty, OpsGenie. Forward every authorization to your SIEM. Subscribe to "cap-in-7-days" or "P0 admit rate < 99%" alarms.

// Mock · tarmac · ops · may 2026
tarmac · ops · fleet view
Admit rate · 24h
96.4%
Holds over-cap
312runs
Active agents
18fleets
// Top 6 by spend · last 7 days
AgentPriTierCap burn
support-triageP0XL62%
revenue-opsP0XL48%
data-pipeline-euP1L71%
code-refactorP2M88%
eval-harnessP2M82%
research-crawlerP3Scap
// 03 · Surface 02

Financial Console. AI compute spend that finally behaves like a managed budget.

// Surface 02 of 02
For the CFO & finance team

Statements. Attribution. Forecast. Audit-ready receipts on every run.

The Financial Console reads from the same ledger as Operations — never a separate "billing" dataset reconciled at month-end. Period statements are GL-ready, with optional one-click export to NetSuite, QuickBooks, Sage Intacct, and Workday.

01

Period statements

Monthly close with full attribution by cost center, team, agent, lane, and provider. PDF, CSV, JSON-LD.

02

GL export

Map Tarmac cost centers to your chart of accounts once; every period exports automatically with the right tags.

03

Forecast to close

Burn-rate projection updates on every authorization, not at month-end. Confidence intervals fold in trend, seasonality, and current pace.

04

Allocation rules

Charge eval costs to R&D, production support to OpEx, internal-tools agents to G&A. Per-tag, per-agent, per-time-window rules.

05

Itemized receipts

Every authorized run gets a permanent receipt — agent, lane, tier, cost, hold-and-release timestamps, the policy rule that decided it.

06

Audit ledger

Append-only, hash-chained record of every authorization, hold, throttle, and settlement. Externally verifiable, retained per your policy.

// Mock · tarmac · finance · period statement
tarmac · finance · close · may 2026
May 2026
Spend · $184,200  ·  Cap · $250K  ·  Saved vs PAYG · $42,100
11 days left
Cost centerSpendShareSaved
Customer support4 agents · P0$48,21026%$11,820
Revenue ops2 agents · P0$41,86023%$9,440
Eng productivity5 agents · P2$36,42020%$8,910
Data pipelines3 agents · P1$31,94017%$6,820
Research2 agents · P3 · throttled$25,77014%$5,110
Period total · governed$184,200100%$42,100
// 04 · The price signal

The Compute Price Signal. A live, source-agnostic price map for every model on every provider.

Every authorization decision needs a current cost. The Price Signal aggregates published rates, observed billing, and customer-negotiated discounts into a single normalized $/token (or $/GPU-second on the GPU lane) — refreshed continuously, audited for drift, and surfaced in the ledger.

// Source-agnostic

Same price abstraction.
Inference lane, GPU lane.

Inference-API ships first. The GPU-rental lane lands later as an adapter — same price signal, same enforcement code path. Adding a new provider is a few hundred lines, not a rewrite.

  • Continuous aggregation across providers and regions
  • Customer-specific contract pricing & volume discounts
  • Token-normalized across model families & reasoning tiers
  • Drift detection — alerts when a published rate changes
  • Hot-reload to the decision engine on every change
// price snapshot · 14:22 utc
ModelTier$ / 1M inΔ 24h
Claude Opus 4anthropicXL$15.00$75 / 1M out
Claude Sonnet 4anthropicL$3.00$15 / 1M out
GPT-5openaiXL$10.00$40 / 1M out−4%
GPT-5 miniopenaiM$0.30$1.20 / 1M out
Gemini 2.5 ProgoogleL$2.50$10 / 1M out−2%
Haiku 4.5anthropicS$0.80$4 / 1M out
// 05 · The core mechanic

Pin the lane. Move the model. One frozen onboarding decision, then per-run optimization within it.

The lane — a model family on a specific provider, or a fixed GPU type — is the part security cares about. We freeze it at onboarding so the security review answers a stable question. What Tarmac changes on every run is the model tier — S / M / L / XL — based on the live price and the remaining budget.

One lane per agent. Four tiers within.

Lane = Anthropic Claude on Bedrock us-east, or OpenAI on Azure swedencentral, or NVIDIA H100 on a fixed cluster. Tiers within that lane are S / M / L / XL — Haiku to Opus, GPT-5-mini to GPT-5, A100 to H200. The decision engine picks the heaviest tier that still fits the remaining budget at the live price.

revenue-opspriority · P0
S · Haiku
M · mini
L · Sonnet
XL · Opus
data-pipelinepriority · P1
S
M
L · fit
XL
code-refactorpriority · P2
S
M · throttled
L
XL
researchpriority · P3
S · floor
M
L
XL
// lighter ← model tier → heavierrecalc · every authorization
// 06 · Integration

Two ways in. One afternoon of engineering work, either way.

Most teams point their existing agent traffic at the Tarmac proxy and are governed inside an hour. Teams that want fine-grained tagging at the call site use the SDK. Both modes hit the same Budget Engine; the receipts are identical.

// 01 · Proxy mode · default

Change the base URL. That's it.

The proxy is OpenAI-, Anthropic- and Bedrock-API-compatible. Point your SDK at api.tarmac.dev; Tarmac authorizes, forwards on a short-lived scoped token, and writes the receipt.

# before
client = Anthropic(
  api_key="sk-ant-..."
)

# after
client = Anthropic(
  base_url="https://api.tarmac.dev/v1/anthropic",
  api_key="tarmac_pk_...",
  default_headers={"x-tarmac-agent": "support-triage"},
)
Time to govern. < 1 hour Code change. Base URL + agent tag
// 02 · SDK mode · for finer attribution

Wrap the call. Tag the work.

For teams that want per-tool, per-step, per-customer attribution baked into the receipt — wrap calls with the Tarmac SDK. Adds a single context-manager; passes through every header and feature of the underlying SDK.

# pip install tarmac
from tarmac import authorize

with authorize(
  agent="support-triage",
  tool="ticket_summarize",
  attribute={"customer_id": cust_id},
) as auth:
  resp = anthropic.messages.create(
    model=auth.model,  # tier-resolved
    messages=msgs,
  )
  auth.settle(resp.usage)
Time to govern. 1 afternoon Attribution. Down to single tool call
// 07 · Roles & permissions

Two surfaces. Four roles. The CFO doesn't see your fleet config. Your engineers don't sign off on the close.

Capability
Platform admin
Platform operator
Finance admin
Auditor
Set budgets & capsper-run · monthly · yearly
r
Onboard agents & lanesset priority tier, configure proxy
r
Edit per-agent rulesthrottle thresholds, soft alerts
r
Close period & export GLstatements, allocations, journals
r
View append-only audit ledgerevery authorization, hold, settle
r
r
r
SSO & SCIM · Okta, Entra ID, Google Workspace · Workspace scope · multi-team Growth & Enterprise · r = read-only with raw query access on the audit ledger
// Take the product for a spin

Run a paid pilot on one team's fleet.

Design partners are onboarded in pairs. Pilot pricing for the first 12 months, direct access to the founder, and the right to shape the product before it sets.

// What you'll have inside an afternoon

From zero to a governed fleet — in under four hours.

  • Proxy traffic routed · 20 minutes
  • Agent priority tiers configured · 30 minutes
  • Budgets & caps encoded · 30 minutes
  • First authorization receipts visible in the ledger
  • First period statement at the next month-end