Financial control plane for AI agents

The financial control plane for AI agents.

Route LLM calls through TokenPilot to attribute spend by agent and customer, protect provider keys, cache repeated calls, and keep an audit-ready usage ledger.

Per-agent and per-customer attributionEncrypted provider key vaultExact response cache with attributed savings
app.tokenpilot.fr / Overview

Agent Spend Overview

Spend, cache savings, and attribution by workspace

7D

Spend this month

$22.01

Cache savings

$5.03

Active agents

5

Spend by agent

customer scoped
ops-summary-bot
$9.20
qa-reviewer
$5.51
invoice-extractor
$5.05
sales-agent-v2
$4.71
support-agent-prod
$3.53
Cache hit2s ago / gpt-4o / support-agent-prod+$0.000840 saved
Platform

Per-agent visibility for real LLM traffic.

TokenPilot sits between your agents and OpenAI or Anthropic so you can attribute cost by agent and by customer, protect provider keys, cache repeated calls, and write an audit-ready usage log.

Per-Agent Attribution

Tag each LLM call with agent_id and customer_id. The dashboard breaks spend, tokens, and cache savings down by agent and by customer.

Provider Key Vault

Store provider keys outside client apps. Keys are encrypted at rest and scoped to each workspace.

Budget Reserve & Settle

Reserve budget before provider calls and settle spend after usage is known, using nanos-level accounting backed by atomic Lua scripts.

Exact Response Cache

Repeated agent calls served from the exact response cache with no provider call. Savings are attributed to the requesting agent.

How it works

Tag once. See everything.

Send three optional headers from your agent code. TokenPilot threads agent and customer identity through routing, cache, audit, and reporting without presenting roadmap controls as active enforcement.

01

Tag

Applications add three optional headers: X-TokenPilot-Agent-Id, X-TokenPilot-Customer-Id, X-TokenPilot-Task-Type.

02

Route

TokenPilot routes the request through OpenAI or Anthropic, reserving budget before the upstream call.

03

Bill

Tokens are settled at nanos precision after the response. Repeated calls can be served from the exact response cache at zero provider cost.

04

Report

Usage logs and audit metadata carry agent and customer identity end-to-end so the dashboard breaks spend down by agent.

Spend Recommendations

Roadmap

Planned recommendations will use workspace usage patterns to suggest budget and cache improvements. Example: "This repeated support workflow may be a good cache candidate."

Control flow

From opaque LLM calls to auditable AI spend.

Scroll to follow one request through TokenPilot — intercepted, attributed, routed, logged, and rolled up.

  1. 01

    Raw request

    An agent sends a normal LLM request.

  2. 02

    Proxy intercept

    TokenPilot intercepts before the provider call.

  3. 03

    Metadata attribution

    Agent, customer, and task metadata travel with the call.

  4. 04

    Control modules

    The control layer checks keys, routing, cache, and budget state.

  5. 05

    Ledger event

    Each request becomes an auditable financial event.

  6. 06

    Dashboard rollup

    Spend and savings roll up by agent, customer, model, and workspace.

Infrastructure

Controls designed for the request path.

The product story is visibility and governance primitives: secure provider keys, attributed cache savings, and an audit-ready usage ledger.

Request-path control

Provider keys, routing, cache checks, budget reserve, and usage logging happen before or during the provider call path.

Cache-hit billing

Exact response cache hits are logged as zero provider-cost events, with savings attributed to the requesting agent and customer.

Audit-ready usage ledger

Workspace-scoped usage logs keep model, token, cost, cache, agent, and customer metadata available for review.

Pricing

Transparent. No surprises.

All plans include a free trial. Enterprise adds security review support and roadmap planning for identity provisioning.

Starter

$0/mo

Individuals and small teams exploring AI cost control.

  • Up to 3 team members
  • 1M tokens / month
  • API Vault (1 provider)
  • Basic analytics dashboard
  • Community support
Most popular

Pro

$49/mo

Growing engineering teams routing OpenAI and Anthropic traffic through one control layer.

  • Up to 20 team members
  • 50M tokens / month
  • Response cache
  • Workspace budget reporting
  • Multi-provider routing
  • Priority support (24h SLA)

Enterprise

Custom

Organizations requiring security review support and workspace governance.

  • Custom usage terms
  • Enterprise security review support
  • Governance roadmap planning
  • Identity provisioning roadmap planning
  • Audit-ready usage logs
  • Dedicated support options

Need a custom contract? Talk to sales

Trust & Security

Built around controls you can inspect today.

TokenPilot focuses on the controls that exist in the product today: encrypted provider keys, workspace-scoped access, structured usage logs, and safe upstream error diagnostics.

Encrypted Keys

Vault-backed

Provider keys are encrypted at rest and never stored in client applications.

Workspace Access

Scoped

Dashboard reads and settings are scoped to the authenticated workspace.

Usage Logs

Structured

Every routed request can be tracked with model, token, cost, and cache metadata.

Safe Diagnostics

Redacted

Upstream error diagnostics redact provider keys, authorization headers, and emails.

Need a security review?

Review the implemented controls and contact us for enterprise security questionnaires.

Ready for agent traffic

Bring AI spend into one auditable control layer.

Route LLM traffic through TokenPilot to attribute spend by agent and customer, protect provider keys, cache repeated calls, and keep structured usage logs.

Per-agent attribution
Encrypted provider key vault
Exact response cache