AICharli customer API

Issue a CSRF token

Create an account

Errors:400403409429503with an error body

Sign in and set the session cookie

Same code as the dashboard's login() function. Checks the username or email and password and sets the session cookie.

Errors:400401403with an error body

Ask for a password reset email

Send email or username. The answer is always the same 200, whether or not an account matches, and mail is only sent when the server has SMTP configured. The link in the mail is valid for 1 hour. Limited to 10 requests per hour per IP and 3 per hour per identifier.

Errors:400403429with an error body

Set a new password with a reset token

Send the token from the reset link and the new password. A successful reset signs out every older session.

Errors:400403500with an error body

Verify the email address with a token

Send the token from the verification link. A token works once and is valid for 24 hours.

Errors:400403500with an error body

Send a new verification email

Sends a new verification link to the signed-in user when their email still needs verification. Limited to 3 per hour per user.

Errors:401403429with an error body

Clear the session cookie

Clears the session cookie.

Current account

Errors:401with an error body

Current account (legacy path)

Existing dashboard JSON route. Prefer GET /api/v1/me, which adds credits.

Errors:401with an error body

Current assistant, plans, and credits

Returns the plans you can claim and your current assistant, if any. Poll until current.status is READY, then open current.launchPath. READY always includes a non-null expiresAt; while the pod is healthy but the billing clock has not started yet, status stays STARTING and expiresAt stays null. While STARTING, current.provisioning reports the pod's own status, when it was claimed, and warmupDeadlineAt, the point by which the claim is fully refunded and the pod released if it still has not come up. Do not call release or upgrade while current.status is STARTING before provisioning.warmupDeadlineAt. FAILED is terminal: the claim could not be set up and error says why. The charge is refunded automatically; a refund that fails is retried every monitoring cycle until it succeeds. It stays visible for about 10 minutes, then current becomes null.

Errors:401403with an error body

Claim Shared or Private

The server picks a free assistant of the requested tier. Send only tier and durationHours. The request returns at once with status STARTING once the charge and assignment commit; the workspace is prepared in the background. Poll GET /api/v1/assistants until current.status is READY (or FAILED, whose charge is refunded automatically). Do not retry blindly: retrying a claim that already succeeded returns 409 with a clear already-held message and never charges twice.

Errors:400401403409503with an error body

End the current session and refund unused time

Ends the caller's current session and refunds unused time. The server finds the session, so the client sends no id.

Errors:401403409with an error body

Add hours to the current session

Adds hours to the caller's current session and charges credits for them.

Errors:400401409with an error body

Move Shared to Private on the same account

Memory stays on the account. Unused shared time is refunded first. The request can take up to ~3 minutes while the private pod is prepared: use a client timeout of at least 180 s. If your client times out anyway, the upgrade still completes; do not retry blindly, instead GET /api/v1/assistants to read the resulting assignment.

Errors:401409503with an error body

Chat with your assistant

Sends messages to your own READY assistant and returns the answer. Needs the session cookie and the x-csrf-token header; the server holds the inference key, so none is ever sent to or returned to the browser. Set stream to true for token-by-token OpenAI-style server-sent events (data: lines ending with data: [DONE]); otherwise one JSON chat completion is returned. Any model, tools or other field you send is ignored: the model is fixed to your assistant's. The inference service raises max_tokens below 4096 to 4096. Errors before the answer starts carry error and code; 409 means no assistant or one that is not READY yet (poll GET /api/v1/assistants), 429 includes Retry-After, 503 is retryable. If a stream breaks after it started, a final data event with an error object and data: [DONE] close it. Closing the connection cancels the request. Nothing is stored; use POST /api/conversations/{id}/messages for history.

Errors:400401402403409429502503504with an error body

Open the managed assistant (browser redirect)

Same URL the live dashboard Open assistant button uses. Requires a session cookie. Returns 302 to the auto-login entrypoint when READY, or an HTML wait page while starting. Use from a top-level navigation or window.open, not as JSON.

Errors:403503with an error body

Resume a stopped assistant runtime

Used by the built-in dashboard. {id} is the internal pod id, which the customer API does not return, so a custom frontend should not call this; poll GET /api/v1/assistants after claim instead.

Errors:401403with an error body

List account conversations

Ordered by updatedAt, newest first, so the conversation with the latest message comes first. updatedAt moves when a message is added (conversations that have had no message since they were created keep their creation time). Returns at most limit (up to 100) conversations, with no paging past that. Metadata only; read a conversation's messages with GET /api/conversations/{id}/messages. Delete one with DELETE /api/conversations/{id}. Admin accounts get 403.

Errors:401403500with an error body

Create a conversation record

Does not start a model reply. A JSON body is required; send {} to create a conversation without a title, which is stored as null. There is no rename yet.

Errors:400401403500with an error body

List messages in a conversation

Newest first: the first message in the list is the latest. Returns the newest limit messages (up to 200); pass order=asc to list them oldest first instead. hasMore is true when older messages exist, and nextBefore is then the id to pass as before to load the next older page (it is null on the last page). before returns the limit messages just before that message, in the same order setting. An unknown before id, or an order other than asc or desc, gets 400. Each message lists its attached files and generated artifacts.

Errors:400401403404with an error body

Store a message

Persists history. It does not call the GPU model. content is trimmed before it is stored and must not be empty after trimming. Leave podId and uploadedFileIds out: the customer API exposes neither pod ids nor file upload yet; both are only used by the built-in dashboard. 403 covers admin accounts, a pod that is not assigned to the caller, and unavailable files.

Errors:400401403404with an error body

Delete a conversation

Deletes one of your conversations together with its messages. Files you attached stay available but are no longer linked to it. 404 covers an unknown id and a conversation that belongs to someone else. Admin accounts get 403.

Errors:401403404500with an error body

Start a credit top-up

Creates a Stripe Checkout session for one credit pack and returns its url; send the customer there to pay. Credits are added once Stripe confirms the payment, never from the redirect back. Answers 503 with code PAYMENTS_DISABLED until payments are enabled, and 403 for admin accounts and for unverified email addresses when verification is required.

Errors:400401403429500502503with an error body

Top-up packs and payment history

Returns whether payments are enabled, the pack amounts you can buy, the currency, and your 50 most recent payments, newest first.

Errors:401with an error body

Auth-proxy health (dashboard poll)

Reports whether the inference proxy is reachable.

Process liveness

Accepts traffic when the database is up

Errors:503with an error body

This OpenAPI document

OpenAI-compatible chat completions

Managed assistant inference. Authorization is a Bearer key minted for the claimed assistant, not a customer-supplied key. Prefer launchPath for the product chat UI.

Errors:401with an error body

List managed models

Search the account's conversation memory