Intencion
Product analytics that improves your AI agents continuously
Last verified:
What is Intencion?
Intencion is a product analytics platform specifically built for AI agents that helps teams ship agents that get better every week. It acts as a product manager for your AI agent by analyzing what worked, what needs improvement, and identifying what to build next. The tool captures every agent run with model, tokens, latency, and a success/failure verdict that you define, not a model's guess.
Key features include three main views: Intents (what users want with success rates), Run Detail (what your agent did with every tool call in order), and Emerging (what to build next - requests the agent can't handle yet sorted by frequency). It auto-instruments with OpenAI and Anthropic, ships issues directly to Linear or GitHub pre-filled with evidence, and works with TypeScript and Python SDKs. The platform groups runs by business intent like refund_request and ranks failures and missing features by how often they happen.
Intencion is designed for product managers, engineering teams, and AI developers building production AI agents who need to track agent performance, fix failures systematically, and prioritize their roadmap based on real user demand. It's ideal for teams building customer-facing AI agents for tasks like order tracking, refunds, account changes, and other business workflows where success rates matter.
Intencion pricing
Pricing model: Freemium
Start free in a minute with no sales call required. The website mentions a free tier but does not display specific pricing details for paid plans on the public website. No call required to start. A live demo is available before signup.
Intencion pros
- Captures every agent run with model, tokens, latency, and custom success/failure verdict
- Auto-instruments OpenAI and Anthropic with single SDK patch
- Groups runs by named business intent like refund_request instead of log lines
- Shows three clear views: what users want, what agent did, what to build next
- Success rate climbs every week as you fix top problems systematically
- Ships issues to Linear/GitHub pre-filled with failure evidence in one click
- Works at class level so framework calls are caught automatically
- TypeScript and Python SDKs available with npm install @intencion/sdk
- Emerging view shows unmet requests sorted by frequency for roadmap prioritization
- Background flush with 5 second delay for efficient data capture
- Run detail shows every tool called in order with exact failure location
- Impact metrics show failed runs per week to quantify fix priority
- No model guess for success/failure - you define the outcome verdict
- Start free in a minute with no sales call required
- Live demo available before signup
Intencion cons
- Requires SDK integration - must patch client code once
- Only supports OpenAI and Anthropic models currently
- Only TypeScript and Python SDKs published, no other languages
- Success rates shown are illustrative in marketing materials
- Requires defining intent labels or relying on auto-inference per run
- Focuses on product layer not traces - may not satisfy deep debugging needs
- No public pricing page visible - free tier exists but paid plans unclear
- Best for agent traffic analysis, not single-agent testing scenarios
Frequently asked questions about Intencion
What is Intencion?
Intencion is product analytics for AI agents that helps you ship agents that get better every week. It acts as a product manager for your AI agent by analyzing what worked, what to improve, and finding out what to build next.
How does Intencion capture agent runs?
You patch your client once with the SDK. Every run is recorded including model, tokens, latency, and a success or failure verdict that you define, not a model's guess. It works at the class level so calls your framework makes on its own are caught too.
What are intents in Intencion?
Intents are named business goals like refund_request. Each run is labeled with the user's goal as a named business intent. You can declare it yourself or Intencion infers it per run from the input. Runs with the same goal are grouped together.
What three views does Intencion provide?
The three views are: Intents (what users want with share of runs that succeeded), Run Detail (what your agent did - every tool it called in order and where it ended up), and Emerging (requests your agent can't handle yet sorted by how often people ask).
How does Intencion help ship fixes?
You can turn any failure or unmet request into a Linear or GitHub issue pre-filled with evidence in one click. The issue includes impact metrics like failed runs per week, where it fails, and suggested fixes like adding OTP fallback.
Which programming languages does Intencion support?
Intencion has published SDKs for TypeScript and Python. You install via npm i @intencion/sdk for TypeScript.
Which AI models does Intencion work with?
Intencion auto-instruments OpenAI and Anthropic models. The SDK includes methods like instrumentAnthropic to wrap your AI client.
How do I calculate success rate with Intencion?
You define the success or failure verdict for each run, not a model's guess. Runs are grouped by intent and success rate is calculated as the share of runs that succeeded for each intent. Each fix raises one intent's success rate.
What is the Emerging view for?
The Emerging view shows requests your agent can't handle yet, sorted by how often people ask. This becomes your roadmap for what to build next, showing new intents like Book a callback with their request counts.
Do I need to talk to sales to start using Intencion?
No. You can start free in a minute with no call required. A live demo is also available if you want to try it first before signing up.