Continue a conversation
Use its history entry or the returned conversation ID. The current request can include bounded recent messages from that owned conversation.
GALACTRE / QUICK START
Try an available model, bring a focused question and inspect the answer. Follow the path that fits you: a guest conversation, your account or a developer integration.
01 / CHOOSE YOUR PATH
Each access path has a different purpose. A public conversation, an account workspace and a server integration do not share the same credentials or permissions.
Use a limited Lunar conversation to explore the text chat experience. No account or browser API key is needed.
Open guest chatOpen GT Chat and choose an available model. Guest chat currently supports Lunar.
Review the guest quota shown by chat. Limits also apply by IP and across the public service.
Send a focused text prompt without credentials or private workspace data.
Stay in the same browser session to revisit guest conversations. Signing in opens separate account history.
Guest availability depends on service configuration, the model catalog and runtime health.
Use the project login or registration form connected to IAMC. The account path provides separate account history and authenticated workspace features.
Sign in to Galactre Create an accountOpen the Galactre login page, or register an account through the project form.
Complete any IAMC authorization or remembered-device PIN step required for your account.
Return to chat for account history; guest conversations do not automatically migrate.
Open Credits and billing only when you want to inspect packages or invoices. A purchase is not required just to read this guide.
Account access, credit balance and permitted models are checked separately. A plan does not activate an unavailable model.
The browser example below uses the Galactre origin, a server session and CSRF. For a backend integration, create an expiring, model-scoped server key in API Access and use the documented responses endpoint.
Read the browser exampleRead the model catalog before selecting a family and verify the integration environment.
For a same-origin browser, discover the session and send the returned CSRF token on mutations.
For another application, verify that a supported Galactre server integration has been provisioned before building against it. Create an expiring, model-scoped key in API Access.
Keep service credentials on the server. Add error handling, budgets, output validation and review before connecting external tools.
The Galactre browser session endpoint is not a general server API-key issuing service.
02 / SEND A FOCUSED REQUEST
Start with a clear question before building a larger workflow. Every example below opens as a draft, so you can edit the text before any request is sent.
Open the current model catalog and select an available family. Release stage, availability, permissions and runtime health are distinct.
Explore model familiesState your goal, relevant facts, constraints and preferred output. Ask for missing information to be identified instead of invented.
Explore instruction recipesSubmit the prompt yourself, then review the reply against your sources. Generated code and test suggestions need independent execution and review.
Learn how to evaluateUse follow-up messages in the same conversation for bounded recent context. New Chat starts a separate conversation without deleting the earlier one.
Understand conversation contextExamples ask for the information needed for a useful answer. Replace them with your own non-sensitive task when ready.
Explain the difference between a PHP queue worker and a synchronous web request. Use a small practical example, identify trade-offs and ask about missing application constraints. Do not invent benchmark numbers.Help me improve a developer-facing release note. Ask me for the original text, intended audience and length before drafting. Preserve technical facts and flag missing information instead of inventing dates or compatibility claims.Help me prepare a focused code review. Ask for the code snippet, runtime version, expected behavior and acceptance criteria. Separate potential defects from questions and suggest tests without claiming to execute them.03 / KEEP THE THREAD
Ask a follow-up in the same chat to build on recent messages. Restate important facts when a conversation gets long, because saved history and included context are different.
Read the context guideUse its history entry or the returned conversation ID. The current request can include bounded recent messages from that owned conversation.
Choose New Chat. That starts a separate conversation and does not delete the earlier one or automatically read other chats.
Guest history is tied to its browser session and expiry. Signed-in account history is separate and guest conversations are not automatically transferred.
04 / YOUR FIRST BROWSER INTEGRATION
This JavaScript example is for a page on the Galactre origin. It checks the model and session, sends a current CSRF token, and accepts only a completed response. Copying the code does not execute it.
/api/v1/chat/modelsRead family availability. Guest chat currently supports Lunar.
/api/v1/chat/sessionDiscover guest/account state, current guest limits and a CSRF token.
/api/v1/chat/generateSend message, family a unique request_key, and an optional owned conversation_id.
The published contract below uses a same-origin Galactre session. For a backend application, issue an expiring model-scoped key in API Access and use the responses API with an Idempotency-Key. Keep browser cookies and CSRF tokens in the browser.
This minimal example requests JSON. Set stream to true for real SSE start, delta and completed events; retain the delivery ID for status and cancellation. Signed-in chat can use selected knowledge sources and bounded tools.
Full authorization guide// Run on the Galactre origin. Calling this function sends a generation request.
async function askGalactre(message, conversationId) {
const catalogResponse = await fetch('/api/v1/chat/models');
if (!catalogResponse.ok) throw new Error('Catalog unavailable.');
const catalog = await catalogResponse.json();
if (!catalog.models?.some(m => m.family === 'Lunar' && m.available)) {
throw new Error('Lunar is not currently available.');
}
const sessionResponse = await fetch('/api/v1/chat/session', {
credentials: 'same-origin'
});
const session = await sessionResponse.json();
if (!sessionResponse.ok || !session.chat_available || !session.csrf_token) {
throw new Error('Chat is temporarily unavailable.');
}
const text = message.trim();
const maxLength = session.limits?.max_message_chars ?? 8000;
if (!text || Array.from(text).length > maxLength) {
throw new Error('Check the message length before sending.');
}
// Retain this key for recovery; a retry with changed input must use a new key.
const requestKey = 'quickstart_' + crypto.randomUUID().replaceAll('-', '');
const body = { family: 'Lunar', model: 'gt-lunar', message: text, request_key: requestKey };
if (conversationId) body.conversation_id = conversationId;
const response = await fetch('/api/v1/chat/generate', {
method: 'POST', credentials: 'same-origin',
headers: {
'Content-Type': 'application/json',
'X-CSRF-Token': session.csrf_token
},
body: JSON.stringify(body)
});
const result = await response.json();
if (response.status === 429) {
throw new Error('Rate limited. Retry-After: ' + response.headers.get('Retry-After'));
}
if (!response.ok || result.status !== 'completed') {
throw new Error('Inspect the request outcome before retrying.');
}
return {
text: result.response,
requestId: result.id,
conversationId: result.conversation_id,
historyState: result.history?.state ?? 'not_reported'
};
}
// Review the returned text. This function does not run code or tools.05 / WHEN SOMETHING NEEDS ATTENTION
Check the actual response before resending. A rejected input, a quota limit and an uncertain execution outcome require different handling.
Enter a non-empty message and wait for any active response to finish. Check the character limit shown by the session; whitespace alone is not a valid request.
Wait for the time indicated by Retry-After or the reset information shown by chat. Avoid repeated retries. Session, IP and public-service limits can apply together.
Refresh the session through the normal chat flow and retry only after checking whether a previous request completed. Account features may require signing in again. Never replace a session token with a provider key.
Recheck the live model list. Use a family marked available and permitted for your session. Guest chat supports Lunar; credits or a URL change do not enable an unavailable family.
The session or model service may be unavailable. Keep your draft and try again later. Repeatedly resubmitting cannot repair service configuration.
Do not assume a lost connection means the request never ran. Inspect history and the returned status before retrying. Treat any downstream operation independently under your own recovery policy.
The completed result may indicate that history storage was unavailable. Preserve important non-sensitive text yourself; do not assume it will reappear after a reload.
Check that you are using the same guest browser session or the correct account. Guest expiry, deletion and the bounded recent-history list can affect what is accessible.
06 / GO A LITTLE FURTHER
Explore model choice, stronger instructions and the boundaries around a useful workflow.
07 / QUESTIONS, ANSWERED
Use gt-lunar to follow the approved Lunar default, or select an exact ID from the releases list for reproducible product configuration. Keep the resolved model_id from each completed response. Review the Models version directory before upgrading.
When guest chat is enabled, you can send limited Lunar text requests without an account or a browser API key. Account history, billing and workspace features require login. Signed-in users can create model-scoped server keys in API Access.
No. Example links open an editable chat draft. Nothing is sent until you submit the message. Usage limits and any applicable consumption are checked when a real request is admitted.
The current chat release accepts text. Signed-in chat provides bounded calculator and current-time tools and selected knowledge sources. File uploads, image generation, arbitrary tools and repository execution are not released. Supply only relevant authorized text within the session limit.
The example is for a page on the Galactre origin with same-origin credentials. Do not assume cookies or CSRF work across another site. A separate application backend needs a supported Galactre server integration; create a model-scoped server key in API Access and use the documented responses endpoint.
Yes. Set stream to true for SSE start, delta and completed events. The start event provides the delivery ID for status and cancellation. Tool planning stays private; the final answer can stream. Check the terminal event before treating delivery as confirmed.
Reuse the conversation_id returned after a saved exchange, with the same authorized identity. Omit the ID for a new chat. A conversation ID alone does not grant access, and bounded recent context is separate from saved history.
It explains the current product and request contract. A production integration also needs verified access, model availability, usage policy, monitoring, recovery, data handling and review of any external actions. Example code does not activate those services.
START WITH A QUESTION
Open chat, choose an available model and send a task you can review.