NullSpend

FinOps layer for AI agents — cost tracking, budget enforcement, and human-in-the-loop approval.

The Problem

AI agents call LLMs autonomously. Without guardrails, a single runaway loop can burn through thousands of dollars before anyone notices. Existing observability tools can alert you after the damage is done. NullSpend prevents it.

How It Works

Your App ──► NullSpend Proxy ──► OpenAI / Anthropic
                │                       │
                │  ◄── streams back ────┘
                │
                ▼
         Dashboard & Webhooks

Two config changes. No SDK. No code rewrite.

# Before                                    # After
OPENAI_BASE_URL=https://api.openai.com/v1   OPENAI_BASE_URL=https://proxy.nullspend.dev/v1

Add an X-NullSpend-Key header and every request is tracked, budgeted, and visible.

Features

Feature	What It Does
Cost tracking	Per-request cost calculation for every model — input, output, cached, and reasoning tokens
Cost attribution	Break down AI spend by customer, team, or feature. Group by API key or any tag, drill into daily trends, export CSV for finance
Budget enforcement	Hard spending ceilings. The proxy returns `429` before the request reaches the provider
Velocity limits	Detect runaway loops — block when spend rate exceeds a threshold within a time window
Session limits	Per-conversation spend caps tied to a session ID
Tags	Attribute costs to teams, environments, features, or anything else via `X-NullSpend-Tags`
Tracing	W3C `traceparent` propagation and custom trace IDs for request correlation
Webhooks	15 event types with HMAC-SHA256 signing — cost events, budget exceeded, velocity alerts, threshold crossings
HITL approvals	Human-in-the-loop approval workflow for high-cost or sensitive operations
Organizations	Team collaboration with roles (owner, admin, member, viewer), per-org billing, and invitation management
Request logging	Opt-in capture of full request/response bodies (including streaming) for debugging and audit (Pro/Enterprise)
Session replay	Group LLM calls by session ID, view the full chronological timeline, expand to see request/response bodies
Multi-provider	OpenAI and Anthropic in a single dashboard with provider breakdown
MCP support	Budget enforcement for Model Context Protocol servers and proxies

Trust Model

The proxy never modifies your requests or responses. Your provider API keys stay with you — they pass through to the upstream provider and are never stored. By default, NullSpend sees only the token counts in the response to calculate cost — prompt content is not logged. Pro and Enterprise plans can opt in to request/response body logging for debugging and audit purposes; bodies are stored encrypted in R2 with per-org lifecycle policies.

Pricing

	Free	Pro	Enterprise
Price	$0/mo	$49/mo	Custom
Proxied spend	$5K/mo	$50K/mo	Unlimited
Budgets	3	Unlimited	Unlimited
API keys	10	Unlimited	Unlimited
Team members	3 (viewers unlimited)	Unlimited	Unlimited
Webhooks	2 endpoints	25 endpoints	Unlimited
Data retention	30 days	90 days	Unlimited
Request logging	--	Full request/response bodies	Full request/response bodies
Key features	Cost tracking, budgets, team orgs	Unlimited keys/budgets/members, request logging	SSO/SAML, custom RBAC, dedicated support

Get Started

Set up cost tracking in under 2 minutes:

OpenAI Quickstart
Anthropic Quickstart
Use with AI Coding Assistants — copy-paste blocks for Cursor, Claude Code, Copilot
Migrating from Helicone

For AI agents: fetch /llms.txt for the complete machine-readable API reference.

API Reference

Build programmatic integrations with the NullSpend API:

API Overview — authentication, pagination, errors, ID formats
Cost Events — ingest and query cost data
API Keys — key management and identity introspection
Budgets — spending limits and budget status
Webhooks — endpoint management and delivery history
Actions — human-in-the-loop approval workflows

SDKs

Client libraries and adapters for integrating NullSpend into your stack:

JavaScript SDK — @nullspend/sdk — TypeScript/JavaScript client for the NullSpend API
Python SDK — nullspend — Python client for the NullSpend API
Claude Agent Adapter — @nullspend/claude-agent — routes Claude Agent SDK calls through the proxy
MCP Server — @nullspend/mcp-server — exposes approval tools to any MCP client
MCP Proxy — @nullspend/mcp-proxy — gates upstream MCP tool calls through approval and budget enforcement

NullSpend

On this page