don't be a
meat proxy.

The handoff

Set the question.
Let your agent read.

You keep the decisions and permissions. AgentFeeds supplies a starting point for research.

  1. 01

    Choose an actual decision

    Start with a collection such as choosing a coding model or planning a platform upgrade. Read its source references and comparison questions.

    Browse the collections ↗
  2. 02

    Give the reading job to your agent

    Use the instruction below, or build a custom reading profile. Save the downloaded profile.json and READING.md in that agent’s project. Do not put secrets in its label or feed links.

    Build a custom reading profile ↗
  3. 03

    Check what it actually found

    Ask for original source links, dates, relevant comparisons and uncertainties. A copied instruction is not evidence that an agent read a source. A release notice is a reason to investigate, not an instruction to change your tools.

  4. 04

    Choose when it looks again

    Ask it to read before a decision, or configure your own harness’s scheduler. A requested reading interval in a profile is a preference. It does not activate a hosted schedule.

    Read the AgentFeeds setup guide ↗
02 / Try a direct routeUses real source collections

Give your agent
a place to start.

Choose a decision. Take this instruction to the agent you already use.

Your agent needs permission to read the web. Its tools, model access and schedule stay under your control.

Agent instructionPlain text
Inspect the sources ↗

Read the sources yourself whenever you want.

Different agents need different reading lists. Customize this reading profile on AgentFeeds →