Developer tools

Connect developer tools and coding agents to BullSequana AI through the BSQAI API.

Agentic Friendly

BullSequana AI can be used from developer tools and coding agents that can be pointed at a custom base URL.

The recommended integration pattern is always the same:

  • point the tool to the BSQAI API
  • authenticate with a JWT or sk-bsq-... API key
  • avoid direct dependence on internal platform components such as KServe

Which tools work

The BSQAI API serves three wire formats:

  • the OpenAI Responses API at https://api.<platform-domain>/v1
  • OpenAI Chat Completions at the same base URL, https://api.<platform-domain>/v1
  • the Anthropic Messages API at https://api.<platform-domain>/anthropic

A tool with an OpenAI or "OpenAI compatible" provider works with the /v1 base URL, whichever of the two OpenAI formats it speaks. Codex CLI speaks only the Responses API and needs wire_api = "responses".

Chat Completions calls a model directly

/v1/chat/completions is stateless: the tool sends the whole conversation and runs its own tool calls. It takes a model, not an Agent, and offers no server-side tools such as file search, web search, or Connectors; those need the Responses API. A request that sets n above 1, asks for log probabilities, or asks for audio is refused with 400. The legacy text completions endpoint, /v1/completions, is not served.

The fastest way to a configuration

The Portal writes the configuration for you. Open a model in Studio → Available models, go to its Use model tab, and pick your agent under Use with a coding agent. The Portal offers Claude Code, Codex CLI, OpenCode, Hermes Agent, Pi and Flue. The install command, the configuration and the run command are shown with this model's id and this environment's base URL already filled in, along with the path of the file the configuration belongs in.

See Studio — Use model.

Available guides

Authentication

For interactive or short-lived usage, JWT can be enough.

For local developer workflows and reusable tool configuration, API keys are usually the better fit.

An API key carries only the permissions chosen when it is created. A coding agent needs:

  • chat.message.create to generate answers through /v1/responses and /v1/chat/completions, and for every route under /anthropic, including its model list
  • tenant.read to list models through /v1/models
  • agent.configuration.use to use an Agent as the model, through /v1/responses

When you belong to a group in the tenant, choose the Team the key's usage counts for. It cannot be changed later, and that team's monthly BCU cap applies to the key as well as your own. A reached cap refuses new requests with 429 usage_cap_reached.

See Create and use an API key.

Start here if you need the authentication flow first:

A coding agent needs tool calling

A coding agent reads files and runs commands through tool calls, and streams its answers. A model deployed without tool calling answers chat but cannot act, so deploy it with tool calling enabled — for vLLM, --enable-auto-tool-choice and a --tool-call-parser matching the model family. The Model Installer adds both to a Qwen, Mistral, or gpt-oss model deployed without a recipe or tool-call arguments of its own, as Tool calling describes.

Tools through MCP

These guides connect a tool to platform models. To give a coding agent access to platform tools instead, point it at a Connector through the MCP Gateway.

On this page