Open source · Hosted beta

Your AI says it read the source. x-reader makes it prove it.

Evidence-aware source reading for AI agents. Give x-reader a public URL and it tells you what was actually retrieved, what is still missing, and how far the answer is allowed to go.

Try x-reader free →View source

The problem

Agents are good at finding something. That is not the same as reading the source.

A search result can look like the article. An X caption can look like the attached video. A YouTube description can look like spoken content. Once those layers blur together, a confident answer can be built on evidence that was never retrieved.

Search snippet ≠ sourcePost text ≠ videoDescription ≠ transcript

Evidence Receipt

Post textPASS

Attached videoPARTIAL

OverallPARTIAL

The post can support claims from its text. The unread video cannot. x-reader keeps that boundary visible instead of silently upgrading it to “read.”

Why it exists

Source verification should be a primitive, not a prompt trick.

x-reader does not ask a model to be more careful. It creates a machine-readable evidence contract the model, workflow or researcher can inspect.

Read the original layer

Start from the URL the user actually gave you, not a search preview that merely mentions it.

Keep missing media missing

If attached video or audio matters but was not consumed, the result stays partial.

Return an evidence receipt

Humans and agents get the same explicit boundary: what passed, what did not, and why.

The contract

Four states. No fake completion.

PASSThe requested source layer is actually evidenced as read.

PARTIALUseful material was read, but a material layer is still missing.

FAILThe requested source could not be fetched reliably.

UNKNOWNThe material is too ambiguous to prove coverage.

How to use it

One evidence model, four product surfaces.

Zero installWeb

Paste a public X, YouTube or web URL and inspect the receipt immediately.

Open app →

For agentsAgent Skill

Add source-first reading to an existing agent workflow.

npx skills add runesleo/x-reader --skill x-reader

For scriptsCLI

Read URLs directly and return structured JSON for automation.

x-reader URL --json

For toolchainsMCP

Expose source reading as a reusable tool inside MCP-capable agents.

Setup →

For researchers

Paste a URL. Check the boundary before trusting the conclusion.

The hosted beta is public-only and deliberately does not inherit browser sessions, Telegram credentials, private cookies or transcription keys.

URL → reader → evidence coverage → PASS / PARTIAL / FAIL / UNKNOWN

For builders

Use the receipt as a guardrail inside the workflow.

Stop treating source retrieval as an invisible preprocessing step. Make evidence coverage inspectable before summarization, analysis, citation or decision-making.

Try a URL GitHub

Try the actual product

Do not ask your agent whether it read the source. Ask for the receipt.

Free hosted beta. No signup. Public URLs only.

Open x-reader →