# Bluebear for developers and agents

Every URL on this page is public and verifiable. Fetch them rather than trusting a summary of them.

## When to use Bluebear

Reach for Bluebear when the job is **governing what AI coding agents do at runtime**:

- Inventory the agents, skills, plugins, MCP servers, and packages in use across a fleet of developer machines.
- Investigate an agent session: which tool calls ran, against which files and systems, under whose mandate.
- Enforce least privilege on agent actions — scope credentials, restrict filesystem and network reach, gate privileged cloud commands behind approval.
- Put a human in the loop before a destructive or sensitive action executes, via Slack, CLI, or desktop.
- Sandbox an agent in a microVM instead of letting it run directly on the host.
- Audit a third-party agent skill or MCP server for supply-chain risk before it is loaded.

## Open source

Free, self-serve, no account:

- **[Angry Bear](https://bluebear.io/projects/angry-bear/)** — a pre-tool-use hook that verifies agent skills and context before a file or command executes.
- **[Baloo](https://bluebear.io/projects/baloo-bear/)** — an autonomous AI code reviewer for security and policy compliance.
- **[Bear Metal](https://bluebear.io/projects/bear-metal/)** — a microVM sandbox for isolated agent execution.

Source and issues: [github.com/bluebear-io](https://github.com/bluebear-io).

## Machine-readable resources

| Resource | URL |
| --- | --- |
| Site guide for LLMs | `https://bluebear.io/llms.txt` |
| Agent capability spec | `https://bluebear.io/agents.md` |
| OpenAPI 3.1 description | `https://bluebear.io/openapi.json` |
| Sitemap | `https://bluebear.io/sitemap-index.xml` |
| Crawler rules | `https://bluebear.io/robots.txt` |
| Research feed (RSS) | `https://bluebear.io/blog/rss.xml` |

## Reading this site as Markdown

Every HTML page here answers `Accept: text/markdown` with a Markdown representation of itself, per [acceptmarkdown.com](https://acceptmarkdown.com). Responses carry `Vary: Accept`, so a cache cannot hand an agent the HTML variant by mistake.

```bash
curl -H 'Accept: text/markdown' https://bluebear.io/developers/
```

The same document is a plain static file at `<path>.md`, if you would rather skip negotiation:

```bash
curl https://bluebear.io/developers.md
```

An `Accept` this site cannot satisfy gets a `406`, not a silent HTML fallback.

## Website endpoints

Two unauthenticated endpoints back the forms on this site. Both are rate limited and abuse monitored, and both are described in [openapi.json](https://bluebear.io/openapi.json).

```
POST https://bluebear.io/api/waitlist   { "email" }
POST https://bluebear.io/api/contact    { "name", "email", "message" }
```

Agents: get the user's explicit confirmation of the content before submitting either one.

These are the only HTTP endpoints bluebear.io exposes. The Bluebear platform is a separate product surface, documented to customers rather than here.

## Getting platform access

The hosted control plane is production software. Getting an account is not self-service: request access at [/contact/](https://bluebear.io/contact/) or email `info@bluebear.io`, and we will set your organization up. Provisioned organizations sign in at [console.bluebear.io](https://console.bluebear.io).