The Baltor deck

1The right setup for every step your agent takes.

Source-backed research and ready files for each step, with setup guides for Claude Code, Codex, OpenCode, Pi and the Baltor Harness. Shows one example step folder.

Notes. Open with the promise, name the agents, then point to the folder.

2Is your AI spending effort in the wrong places?

Too much context for a small task. An expensive model for every decision. Expertise left on the shelf. Code written again. The same mistakes, again. Long work that outgrows a small model.

Notes. Ask which of the six the room recognises. Each costs tokens, time or quality, and each has an answer later in the deck.

3Reusable files for technical and creative work.

Instruction files, skills, tools and reusable code, data and workflows, prompts and task briefs, MCP server settings and plugins. Your models and keys stay yours, and automated checks of licence, source, safety and declared effects run before a file is published.

Notes. These are formats agents already read. Library access and model costs stay separate.

4Search returns short results. A step downloads only what it chose.

Search: the agent asks the library for what a step needs. Choose: the step picks one result. Download and check: access is checked again at download. Place: the file goes where the agent reads it.

Notes. Search returns references; the body arrives with one checked download.

5A working library.

1,560,856 files, searched and downloaded from the agents people already use.

Notes. Everything here is live today.

6Measured in the open.

143 synthetic rows, 72 counted steps, 3 runs per step in a fresh Pi process, 4 task families scored independently. Savings depend on the task, the model and the setup.

Notes. The full report is public, every result included. Read the data cleanup study

7The served library and its growth targets.

Observed: 1,560,856 files and 256,380 components. Plan: two million served files. Aim: one million useful components.

Notes. The two observed figures were read from the service on October 25, 2026; the other two are targets.

8Library access, with your own model provider.

Feeds, $4.99 a month, free through December 31, 2026 (Eastern). Pro, $29 a month, adds Skills and agent files. Max, $49 a month, adds Overnight / AFK Work in Public preview. Payments run through Stripe; Public Good files are free with an account.

Notes. Customers bring their own model access. Paid plans start when you opt in.

9The formats became standards. Choosing and trusting the right file is the hard part.

Agent Skills became an open standard on December 18, 2025. Curated skills raised the mean pass rate from 33.9 to 50.5 percent on 87 tasks in SkillsBench (June 2026). A February 2026 study of 98,380 skills from two registries confirmed 157 as malicious.

Notes. Baltor's answer: one library, delivered one step at a time, with each file's licence and exact version checked before it is used.

10Talk to us.

Create an account, then connect your agent with the guide for the agent you use. Baltor.AI, info@baltor.ai, United States.

Get startedConnect your agent

Sources on the slides

• Baltor capabilities recordSlides 5 and 7
• Data cleanup studySlide 6
• Agent SkillsSlide 9
• SkillsBench, arXiv, June 2026Slide 9
• Skills study, arXiv, February 2026Slide 9

Take the deck with you

The PDF keeps every slide and its notes. The reader lives at baltor.ai/deck, and its address opens on slide 1.