See where every run failed, and why.

Intencion reads any agent the same way: the goal, every tool call, and the step that broke.

Capture every run as a trace, find the failing step, and ship a fix every week.

Start freeRead the docs →

Need help? Get an onboarding session with the founders.

agent.ts
import { Intencion } from "@intencion/sdk";

const ix = new Intencion({ apiKey });
const openai = ix.instrumentOpenAI(new OpenAI());

// every call now traced: model, tokens,
// latency, outcome. Zero per-call code.

One line. OpenAI & Anthropic, TypeScript & Python.

The traceread the failing run

Open the run, see the step that broke.

Coding, data, RAG, support: Intencion reads any agent the same way. The goal, every tool call, and exactly where it went wrong.

read_repo · fs · 0.1s
apply_patch · editor · 0.3s
run_tests · ci · skipped
✗ tool call rejected: required arg "path" missing
Run detail

The exact step that broke

Replay every tool call, in order, with inputs, outputs, and latency. The broken step is marked: a malformed tool call, an ungrounded answer, a step that silently did nothing.

Apply a code patch71%
Answer from our docs84%
Generate a SQL report90%
Resolve a support ticket95%
Outcomes

Which goals are failing

Every run carries a pass-or-fail verdict you define in code. Goals sort by failure rate, so the biggest problem sits at the top of the list.

Migrate a DB schemanew142
Summarize a recordingnew73
Cancel a subscriptionnew41
Unhandled

What it can't do yet

Requests your agent has no path for, ranked by how often people hit them. The failures of absence, and your build-next list.

The loopcapture, ask, fix

From a failed run to a shipped fix.

Patch your client once. The rest is one loop, on repeat.

i.

Capture

Patch your client once. Every run is recorded: each tool call, model, tokens, latency, and a success or failure verdict you define in code.

ii.

Trace

Each run becomes a readable trace, grouped by the goal it was serving. The runs that failed rise to the top.

iii.

Ask why

Open the failing run, or let your coding agent ask Intencion why(run) over MCP. Back comes the failing step, the error, and a fix.

iv.

Fix & ship

Apply the fix with the failing step and its exact error in hand, then ship it.

v.

Watch it drop

The failure rate falls. Next run that breaks, you do it again.

The fixship with the error in hand

Don't just find the bug. Ship the fix.

Every failing goal comes with the step that broke, the exact error, and what to change.

failure · 61%
Change my plan
43% of failures: can't verify identity
→
Change my plan succeeding at 61%, can't verify identity
Impact160 failed runs/wk (~96 tickets)
Wherefails at verify_user (auth), no fallback
☐ success ≥ 85%☐ add OTP fallback
Setupone line, both SDKs

Patch your client once.

Every call is captured at the client level: tool calls, tokens, latency, and the outcome you define. TypeScript and Python.

$npm i @intencion/sdk
Read the docs →
agent.ts
import { Intencion } from "@intencion/sdk";
import Anthropic from "@anthropic-ai/sdk";

const ix = new Intencion({ apiKey });
const anthropic = ix.instrumentAnthropic(new Anthropic());

// every call captured: model, tokens, latency, outcome.
tools.ts
await ix.run({ intent: "refund_request", input }, async (run) => {
  const order = await run.tool("lookup_order", "orders-db", () => lookupOrder(id));
  return await issueRefund(order); // returns → success
});

Stop guessing why your agent failed.

Capture every run, find the step that broke, and ship the fix. Start free in a minute.

Start freeRead the docs →