Bibha BRAHMA-30B is Bibha’s self-hosted, whitelabeled 30-billion-parameter large language model. It runs entirely on your organization’s own GPU infrastructure — no request ever leaves for an external model provider. You send a standard chat-completions request and BRAHMA serves the response, metered per organization. BRAHMA speaks the OpenAI chat-completions dialect, so most OpenAI SDKs and tools work against it by changing only the base URL and API key.

Base URL

BRAHMA is served under the PeachDesk API. Every route below is relative to your PeachDesk host.
The staging base URL is api.stage.peachdeskai.com. The production base URL api.peachdeskai.com applies once production DNS is live.
The chat-completions endpoint is POST /api/v1/voice-engine/llm/v1/chat/completions. Self-hosted deployments use the same path on their own host: https://your-peachdesk-instance/api/v1/voice-engine/llm/v1/chat/completions.

Authentication

BRAHMA requests authenticate with your PeachDesk API key. Pass it in the X-API-Key request header — the same key used everywhere else in the API (see Authentication).
API keys are scoped to an organization, so every completion is served by that organization’s BRAHMA deployment and every token is metered against that organization. A missing or invalid key returns 401; a valid key without access to the resource returns 403.

OpenAI compatibility

The request and response bodies follow the OpenAI chat-completions schema. You can point an existing OpenAI client at BRAHMA by setting the base URL to https://api.stage.peachdeskai.com/api/v1/voice-engine/llm/v1 and authenticating with your PeachDesk key. The full request surface — message roles (including developer and tool), sampling parameters, max_completion_tokens, function calling (tools/tool_choice), structured outputs (response_format), stored completions (store/metadata), and streaming — is documented on Chat Completions. Set stream: true to receive tokens incrementally as Server-Sent Events. Every non-streaming response includes a usage object with prompt_tokens, completion_tokens, and total_tokens.
Requests route to your organization’s self-hosted BRAHMA-30B GPU, never to an external LLM provider. The model field is accepted for OpenAI compatibility; completions are always served by BRAHMA-30B.

Endpoints

Chat Completions

Send an OpenAI-compatible chat-completions request, with optional streaming, and receive a completion from BRAHMA-30B.

Function Calling

Define tools BRAHMA can call, handle tool-call turns, and stream tool-call deltas.

Structured Outputs

Guaranteed-schema JSON via grammar-constrained decoding with response_format.

Stored Completions

Persist completions with store: true, then retrieve or delete them by id.

Prompt Engineering

System-message structure, few-shot patterns, and latency-conscious prompting for BRAHMA.

Models

List and retrieve the models served by the BRAHMA endpoint, in the OpenAI models format.

Rate Limits

Per-organization RPM and TPM limits, rate-limit headers, and backoff guidance.

Errors

The OpenAI error envelope, status codes, and request IDs returned by BRAHMA routes.

Token Usage

Read your organization’s BRAHMA token usage — prompt, completion, and total.

Cost

How BRAHMA usage is metered and priced under the managed credit-token model.

Authentication

Create and manage the PeachDesk API keys that authorize BRAHMA requests.

Quickstart

Send a streaming chat-completion to BRAHMA-30B.
Omit stream (or set it to false) to receive a single JSON response with the full completion and a usage object. Track cumulative consumption for your organization with GET /api/v1/voice-engine/llm/usage (see Token Usage).