AstralBeam

Which LLM providers can I connect through the dashboard, and can I bring my own API keys instead of paying AstralBeam's token rates?

Yes. You can bring your own API keys for OpenAI, Anthropic, Google Gemini, xAI (Grok), and OpenRouter, and those model providers bill you directly for tokens; AstralBeam charges only its own usage-based platform fee plus conversation and file storage. The homepage stack section also lists Vercel AI Gateway, though the site notes it is still in progress. This is covered in the homepage FAQ, the Works With Your Stack section, and the Models docs at /docs/dashboard/models.

Found on / › ADD AN AGENT<br>TO YOUR APP<br>IN MINUTES › FREQUENTLY ASKED QUESTIONS

What does AstralBeam actually cost per month, and is there a free tier I can use while prototyping before committing?

AstralBeam has no fixed monthly plan: AstralBeam Cloud charges $0.20 per million tokens (same rate for input and output) plus $0.50 per GB per month for conversation and file storage, with model and sandbox providers billing you directly. For prototyping you get $20 in free credits every month, which cover AstralBeam fees only, and self-hosting is free of platform and storage fees if you run it on your own infrastructure. Managed hosting on your infrastructure adds a custom hosting fee on top of the same usage rates; these figures appear in the pricing section and FAQ on astralbeam.ai.

Found on / › ADD AN AGENT TO YOUR APP IN MINUTES › PAY FOR WHAT YOU USE › Cloud

How do I register custom tools or widgets the agent can call, and can those tools hit my private backend APIs on the user's behalf?

Yes. You register tools by adding actions to the chat's `tools` prop and widgets via the `widgets` prop, each with a description, a parameter schema, and an execute or render function, as documented on /docs/sdk/tools-and-widgets and demonstrated in the Connect tools & widgets section of the homepage. Those tools run in your page and can call your private backend APIs using the user's session, behind your own permission checks, per the security model at /docs/sdk/security and the FAQ. The Todos tutorial at /docs/start/todos-tutorial walks through a complete working example with shared state and two-way updates.

Found on /docs/sdk/tools-and-widgets › Tools and widgets › Tools

How does SDK authentication work for my end users — do they get their own tokens, how are sessions validated, and how are tokens refreshed or revoked?

Yes. Each end user gets their own short-lived tenant JWT, valid 60 to 600 seconds with a 300 second default, which your server mints from your organization API key at POST /api/astralbeam/token using user and tenant IDs taken from your authenticated server session; the browser never sees the API key (see /docs/sdk/authentication and /docs/sdk/security). Sessions are validated by sending the in-memory token as an Authorization: Bearer header with no cookies, and the SDK synchronizes the signed tenant and current user through /me after every token acquisition. Tokens renew automatically before expiry, and the docs state that deleting a TenantUser does not revoke its token, so the short expiry is what bounds a token's life rather than any revocation endpoint.

Found on /docs/sdk/security › Security model › Tokens

Can I run multiple agents with different system prompts and tool access from one account, and how do I route conversations to the right one?

Yes: the Agents docs (/docs/dashboard/agents) say owners and developers can create and edit multiple agents, each with its own system prompt set in the dashboard (browser-supplied prompts are rejected) plus its own model, web access, and sandbox provider. Route conversations by passing the agent's public ID, agent_<organizationId>_<id>, as agentId; conversations that omit agentId fall back to the organization's default agent, and existing conversations keep their assigned agent. The API reference (/api/openapi.json) confirms each request uses the conversation's selected agent and rejects client-supplied systemPrompt values.

Found on /docs/dashboard/agents › Agents › The system prompt

robots.txt

GET /robots.txt → 200 text/plain (214 characters)

sitemap.xml

sitemap found at /sitemap.xml

llms.txt

/llms.txt has an H1 and 2,867 characters of text; first link https://astralbeam.ai/ → 200 text/html

AI crawler access

robots.txt names no AI crawlers and does not block all crawlers from /

Content signal

robots.txt has Content-Signal: search=yes, ai-input=yes, ai-train=yes

Pages load cleanly

45 of 47 crawled pages returned content (95.8%; needs over 95%)

Server-rendered text

raw homepage HTML has 11,160 visible characters before JavaScript runs (needs 500)

OpenAPI spec

found /api/openapi.json → 200 application/json

Canonical URL

homepage canonical: https://astralbeam.ai/

Meta description

meta description is 151 characters

Page language

<html lang="en">

Organization schema

homepage JSON-LD includes an Organization with a name

Payment docs

site category is not commerce

Webhook docs

site category is not commerce

Auth docs

site category is not commerce

AI agent accessPass

robots.txt allows all 10 search and user AI bots on /. training and mixed-use tokens (0 points): all 6 allowed

Reachable by agentsPass

SitedexBot homepage fetch (/) → 200 text/html; no challenge

Real 404sFails

GET /sitedex-probe-404/e8b1c4d2/missing (a page that does not exist) → 200 text/html: a soft 404 (success status for a missing page)

Return a real 404 status for pages that do not exist, not a 200 page that says “not found” or a redirect to your homepage. Agents read the status code to tell a missing page from a real one.

Link headersFails

homepage response has no Link header (HTML <link> elements are not counted)

Add a Link header to your homepage response that points agents at your machine-readable files, for example Link: </llms.txt>; rel="describedby"; type="text/markdown".

Open Graph tagsPass

homepage has all four required Open Graph tags (og:title, og:type, og:image, og:url)

Organization contact detailsFails

Organization JSON-LD has neither contactPoint nor address

Add contactPoint and address to your Organization JSON-LD if you want agents to find how to reach you. Both are optional in schema.org; leave out an address you keep private.

JSON-LD parsesPass

2 of 2 JSON-LD blocks on the homepage parse as JSON

About, contact and privacy pagesPartly passes

no about page crawled; no contact page crawled; privacy /privacy (11,004 bytes) (needs 500+ bytes of text each)

Publish an about page, a contact page and a privacy page with real text (about 500 characters of text or more each), and link them from your site so a crawl finds them.

Agent guidance in llms.txtFails

/llms.txt has no section that says when to use the site or gives agent instructions (heuristic: an H2-H6 heading such as “When to use” or “Instructions for agents”, with text under it)

Add a section to /llms.txt that tells an agent when to use your site, for example a “## When to use” heading with a few lines on what the site is for and what it is not for.