Features

The full stack of Agent Unit Economics.

Track every cost. Attribute every dollar. Recover the retry waste. Built for teams running AI agents in production.

Track

Capture every cost automatically

One line of code captures LLM calls, tool APIs, compute, and retry waste — across LangChain, CrewAI, OpenAI Agents SDK, or your own loop. No wrappers. No missed costs.

LLM auto-instrumentation

OpenAI, Anthropic, Gemini, Bedrock, Cohere, LiteLLM, Vercel AI — seven providers, zero config. Every token counted, every cost captured.

OpenAIAnthropicGeminiBedrockCohereLiteLLMVercel AI

163-service catalog

Tavily, Firecrawl, Pinecone, Google Maps, E2B, Browserbase... costs extracted from API responses automatically.

Session auto-grouping

No wrapper code needed. Works with LangChain, CrewAI, OpenAI Agents SDK, custom code. Tasks group automatically.

Dev Mode

See every event in your terminal as it happens. Zero cloud required.

Works with your stack

Initialize once. Every LLM call and HTTP request in your agent is captured automatically. No changes to your existing code.

LangChain & LangGraph

CrewAI

OpenAI Agents SDK

Vercel AI SDK

Custom agent loops

import dexcost

# One line to start tracking
dexcost.init(api_key="dx_live_...")

# Track a task with full attribution
with dexcost.task(
    task_type="resolve_ticket",
    customer_id="cust_123",
    project_id="support-bot",
) as t:
    # All LLM calls and HTTP requests
    # inside this block are tracked
    result = agent.run(ticket)

Analyze

Recover spend, spot anomalies, understand margins

Tracking is the input. Find the retry waste. Spot the unprofitable customer. Decompose a cost spike before it becomes a budget incident.

Retry cost recovery

See exactly how much you spend on retries — by agent, by task type, by customer. Identify the expensive loops and recover wasted spend.

Anomaly detection

Automatic alerts when task costs spike. Catch runaway agents before they drain your budget.

Customer profitability

Which customers are profitable? Which are costing you margin? Visualize your cost distribution and find the outliers.

Cost decomposition

What's driving cost changes period over period? Is it more tasks, more tokens per task, or price changes? Break it down.

Attribute

Know exactly who costs what

Full attribution to customers, agents, and task types. Build pricing, find inefficiencies, generate usage statements.

Per customer

Total cost to serve each customer. Revenue minus cost equals margin — per customer, per month.

Per agent

Which agent is expensive, which is efficient? Compare agents head-to-head on cost per task.

Per task type

Average cost per resolve_ticket vs generate_report. Understand the economics of each workflow.

Usage statements

Shareable cost reports per customer per period. Export as CSV or PDF. Ready for billing conversations.

Four SDKs. One schema.

Python, TypeScript, Go, Rust — all instrumenting the same Standard Event Schema. Your data is portable from day one. MIT licensed.

Python

pip install dexcost

TypeScript

npm install @dexcost/sdk

go get github.com/dexwox/dexcost-go

Rust

cargo add dexcost

import dexcost

# One line to start tracking
dexcost.init(api_key="dx_live_...")

# Track a task with full attribution
with dexcost.task(
    task_type="resolve_ticket",
    customer_id="cust_123",
    project_id="support-bot",
) as t:
    # All LLM calls and HTTP requests
    # inside this block are tracked
    result = agent.run(ticket)

7 LLM providers163 services4 SDK languages

Ready to see your true agent costs?

One line of code. 7 providers. 163 services. Full cost clarity.

Start building free Read the docs

Free up to 5,000 tasks/month · No credit card required