botanu Documentation

botanu is the AI cost intelligence platform. It captures every LLM call, tool invocation, and workflow execution in your AI application — then turns that raw telemetry into actionable cost-per-outcome metrics.

If you’re building AI-powered products, you need to answer: “What does it cost to produce a successful outcome?” botanu gives you that number, broken down by workflow, customer, model, and time period.

Why botanu?

Most observability tools show you traces. botanu shows you business outcomes tied to cost.

What you get	How it helps
Cost per successful outcome	Know exactly what it costs to resolve a ticket, summarize a document, or complete any AI workflow
Cost per failed outcome	Identify wasted spend — money spent on runs that never produced value
Workflow-level dashboards	Compare cost, latency, and success rate across all your AI workflows
Span-level drilldown	Inspect every LLM call, tool call, and data lookup within a single run
Retry cost attribution	See the total cost of retries that led to a successful outcome
Customer economics	Understand unit economics per customer across your AI features

How it works

Your App  →  botanu SDK  →  ingest.botanu.ai  →  Dashboard
(2 lines)     (auto-      (OTLP collector)      (app.botanu.ai)
               instruments
               LLM calls)

Install the SDK — pip install botanu or npm install @botanu/sdk
Set BOTANU_API_KEY in your environment
Wrap your agent — one botanu.event(...) block around the entry point provides event_id, customer_id, and workflow
(Optional) Annotate with diagnostics — botanu.emit_outcome(...) stamps value_type, reason, error_type, etc. onto the run. Event outcome itself is resolved server-side from eval verdict / HITL / SoR
View dashboards — cost, latency, and outcome metrics appear in real-time at app.botanu.ai

LLM calls to OpenAI, Anthropic, Cohere, Google, Mistral, and Groq are auto-instrumented inside the event scope — no extra code.

The data model

botanu organizes your AI operations into four levels:

Workflow → Event → Run(s) → Spans

Workflow: A category of AI operation (e.g., “Customer Support”, “Doc Summarization”)
Event: One business unit of work, identified by event_id (e.g., ticket TKT-1234)
Run: One execution attempt. Retries create multiple runs under the same event
Span: One operation within a run — an LLM call, database query, or tool invocation

Learn more in Concepts: Workflows & Events.

Quick links

Quickstart — get traces flowing in 5 minutes
Python SDK — installation, decorators, outcomes
TypeScript SDK — installation, wrappers, outcomes
Existing OTel / Datadog setup — run botanu alongside your existing APM
Content Capture — enable prompt/response capture for eval
AI-Assisted Setup — use Claude Code, Cursor, or Copilot to instrument your project automatically
Concepts — understand workflows, events, runs, and spans
API Keys — create and manage your API keys
Cost Model — how botanu computes model, data, and infra costs

Supported languages

Language	Package	Status
Python	`botanu`	General availability
TypeScript / Node.js	`@botanu/sdk`	General availability
Java	—	Coming soon
Go	—	Coming soon

Open standard

botanu is built on OpenTelemetry. Your traces use standard OTLP format, W3C TraceContext propagation, and W3C Baggage. There’s no vendor lock-in — if you stop using botanu, your instrumentation still works with any OTel-compatible backend.

Quickstart