Have an AI agent? Let it explore Baltor for you. Any assistant works, personal or coding.

Connect your agent
What your agent reads

The request it copies

Explore baltor.ai for me and report back. Start with https://baltor.ai/llms.txt, then read https://baltor.ai/.well-known/agent-card.json and https://baltor.ai/.well-known/mcp.json. Tell me which plan (Feeds at $4.99/mo, Pro at $29/mo, or Max at $49/mo) fits my work, what you would connect, and the setup steps at https://baltor.ai/connect. Stay read-only: read public pages and agent files, and check with me before any sign-in, purchase or change. If a step needs a key, name it (BALTOR_SERVICE_TOKEN) and I will add it myself.

Discovery files on baltor.ai

Read-only discovery is open. Keep BALTOR_SERVICE_TOKEN in your own environment.

Baltor
Sign in Get started

Know what you share. Control what runs.

Baltor serves checked files and research to the harness you run. Your model keys, your project and every permission stay with you.

Two kinds of access

Your Baltor key or OAuth connection reaches the library. Your model access stays in your own tools.

Files wait for your agent

A downloaded file is material to read. Your harness decides what runs, with the permissions you choose.

Exact bytes, every time

Each download names the SHA-256 of the exact bytes, and the service checks the body against it.

Effects need authority

A file that writes, runs commands or reaches the network says so before download, and the step declares the same effects.

Service access and model access are separate.

A Baltor key gives access to the library. Model calls happen in your own tools, through your provider account or your own models.

Baltor access

Who holds it
You: one Baltor key for each tool or device, or an OAuth connection you approve in your browser.
What it reaches
Search, downloads, your usage record and billing sessions, within the scopes you allow.
Where it lives
BALTOR_SERVICE_TOKEN in your environment. The configuration names the variable.

Model access

Who holds it
You: a provider key, Ollama Cloud (https://ollama.com/v1) or an OpenAI-compatible cloud endpoint.
What it reaches
The models your harness calls for each step.
Where it lives
Your environment or an approved credential manager. A remote model receives whatever your tools send it.

Your agent decides what runs.

With the permissions you choose. A file arrives as material to read.

Search returns metadata your account may read. Fetching a body rechecks your access and the item's exact identity. A digest confirms the bytes match the reference you selected; whether they are safe, correct and useful for your task is for you and your harness to judge.

  • Inspect unfamiliar code and plugins before a step uses them.

  • Run them in a confined workspace with explicit file, command and network permissions.

  • Give separate work separate containers. Processes in a shared container can reach one another.

Each step states what it may do with Baltor-Step-Effects. Reading files is the default; a file that declares more effects downloads once the step declares them too.

reads_fswrites_fsreads_secretnetworkspawns_process

How review works

Every item reaches your agents with a record of who approved it, against which criteria, and which checks ran.

The library serves two tiers, Verified and Community. Searches include both by default and can keep to verified items.

Library tiers

  • Every admitted item has an approval record tied to its exact bytes. The record names each reviewer, its decision and its reasons.
  • Reviewers judge items written by others. One written rejection keeps an item out, and the reason is recorded.
  • Full review of one item takes approval from at least two reviewers from at least two model families, each from a family other than the one that wrote the item.
  • An initial screen uses one calibrated, independent reviewer and written criteria: the practice is sound, claims about the cited source are true, the licence is settled and every effect on your computer is declared.
  • Since October 5, 2026, a generated item is published once it passes the automated checks. Sampled review by a different model family continues after publication and withdraws what it rejects, and deliberately defective examples test the reviewer.
  • Automated checks look at the licence, which must be accepted and match the item; completeness; safe instructions, where piping a download into a shell or reading a credential file fails; declared effects against the steps; credential-shaped values; and copies of other items. Material from outside sources carries a provenance record.
  • Each item's evidence records which checks ran and whether independent review is complete or ongoing.
  • The first 43 items were approved on September 21, 2026, before the model-family rule. Three reviewers judged 49 items, approved those 43 and rejected six; one of them shared a model family with the model that wrote the items.
  • A digest proves the downloaded bytes are the approved bytes. Approval covers those bytes; installing, running and judging fit for your task stay with you.

What reaches this service

What you search for, what you request and what you choose to send.

Read the privacy notice

When you use the workspace
Your search text, the references you request and your sign-in. The service records access and usage metadata.
What you choose to send
Feedback, reports and files, through submission and upload features. Send content you have permission to provide, check its audience and retention, and leave out credentials and unrelated personal data.
What stays on your machine
Project files, model keys and execution traces. The website sends what you choose to send.
Embeddings and derived features
They can still identify their source, so treat them like the data they came from.

Current operational limits

Plan sensitive work around these facts, read from the service on October 25, 2026.

Every limit in the record

Where it runs
The capabilities record reports the runtime cloudflare-workers-d1-r2.
Availability
The service can change or be interrupted, as section 2 of the terms of service says.
Checks
The checks on this page are Baltor's own.
Sign-in and keys
Email address and password. Keys are created and revoked on your account page; keys search and download, and your account, billing and keys are managed in your account.
LimitValue
Request size64 KiB
Response size16 MiB
One download1 MiB
Operations at once, per account4
Refused sign-in attempts, per address30 in 60 seconds

Before sensitive or consequential work

  • Read the privacy notice for what the service stores. Privacy notice

  • Set each task's permissions in your own harness.

  • Keep provider credentials out of prompts and retrieved files.

  • Reconcile an uncertain outcome before repeating an external action.

Checked files, your controls.

Pro: Agent Feeds + Skills and agent files, $29 a month. Cancel any time from your account page.

Security questions go to Baltor.AI at info@baltor.ai. Support