Give your agent sources it can check.

What your agent can decide with a feed

19 collections in 4 groups: Compare models, Build and operate, Creative and 3D, Find opportunities. Six of them, with the question each answers:

• Choose a coding model

  Which model and harness combination should handle this repository task?

  Sources: SWE-bench, Stanford HELM

  Open the collectionMarkdown link

• Compare model cost and latency

  Which provider and model meet the workload's quality, latency and spending constraints?

  Sources: Artificial Analysis methodology, Anthropic API pricing

  Open the collectionMarkdown link

• Choose MCP tools and services

  Which service supplies the required capability with acceptable permissions and maintenance?

  Sources: Official MCP Registry, GitHub Releases API

  Open the collectionMarkdown link

• Plan platform and tool upgrades

  Which announced changes require action in this stack, and which can wait?

  Sources: GitHub Releases API, GitHub Changelog, Cloudflare Changelog

  Open the collectionMarkdown link

• Find a fitting hackathon

  Which event fits our project, eligibility and ability to publish the intended demonstration?

  Sources: Official Kaggle CLI, Devpost hackathons

  Open the collectionMarkdown link

• Choose a 3D or game engine

  Which engine or tool should a creative agent build this project on?

  Sources: Godot Engine documentation, Godot download archive, Blender release notes and 2 more

  Open the collectionMarkdown link

Set it up in five steps

1. Choose what each agent decides

   Open Feeds and pick one to three collections for each agent's job.

   You see 19 collections in 4 groups, each with its question, its sources and two links.

2. Copy the link in the format it reads

   Markdown suits instruction files and chat agents. JSON Feed suits scripts and tools.

   You see an address like /feeds/collections/coding-benchmarks.md.

3. Add the link to the agent's instructions

   One line in AGENTS.md, CLAUDE.md or your agent's own instructions file:

   Before choosing a coding model, read https://baltor.ai/feeds/collections/coding-benchmarks.md and follow its decision checklist.

4. Give each agent only its feeds

   Each link stands alone, so different agents and different harnesses read different sources:

   • A coding agent: Choose a coding model and Standardize a coding agent version
   • A research agent: Turn papers into practical tests and Review RAG and retrieval quality
   • A creative agent: Choose a 3D or game engine and Connect creative tools through MCP

5. Decide how often it reads

   Your harness owns scheduling; Baltor does not push. Once a day, or before each decision, is enough. Send the last ETag in If-None-Match and an unchanged feed answers 304 Not Modified:

   ETAG=$(curl -sSI https://baltor.ai/feeds/collections/coding-benchmarks.md | awk 'tolower($1)=="etag:"{print $2}' | tr -d '\r')
   curl -sS -o /dev/null -w "%{http_code}\n" -H "If-None-Match: $ETAG" https://baltor.ai/feeds/collections/coding-benchmarks.md

   You see 304 while the feed is unchanged. Each file names the date its sources were checked and a digest of the directory, so you can tell exactly what an agent read.

Build a reading profile for each agent: choose collections, include or exclude individual sources, and set a reading interval. Download the profile and instructions into that agent's project. Your harness owns the schedule and the saved files.