Bluwing/Docsv0.9.0
Operational
API ReferenceConsole

Get started

  • Introduction
  • Quickstart
  • How it works

Build

  • Tool format
  • Matching hints
  • Toolsets and builds
  • Versions and live

Call

  • Resolve a request
  • Reading responses
  • OpenAI and Anthropic compatibility

Account

  • Authentication
  • Credits and billing
  • Limits
  • Errors
  • Security and data
  • FAQ

Get started

Introduction

What Bluwing does, where it sits in your agent, and what you need to start.

Bluwing picks the right tools for a request, in the right order, and fills in their arguments from what the user said. You pass your prompt and the tool definitions your agent already has. It gives back deterministic, execution-ready calls.

It replaces the brute-force step where you send dozens or hundreds of tool definitions into an LLM context window on every conversational turn. A call to Bluwing takes tens of milliseconds (18–35 ms), costs $0.015 per 1,000,000 input tokens (output tokens are 100% free), and never hallucinates an argument the user didn't provide.

The hierarchical decision tree narrows thousands of tools in sub-milliseconds

#How It Works in Your Architecture

Your agent keeps doing everything it does today. Only the tool-selection step moves to Bluwing:

01
User Prompt

User speaks or sends text to your application

↓
02
Bluwing Gateway

Evaluates prompt and tools using model: "bluwing"

↓
03
Needle 3 Resolution

Traverses compiled tree in sub-milliseconds (<35ms)

↓
04
Deterministic Tool Calls

Returns structured JSON with confidence, arguments, and grounding

↓
05
Your Backend Execution

Runs the chosen functions with your own credentials and databases

  1. User Request: A user talks to your app or agent.
  2. Unified Call: Your server sends the user's text and tool array to POST /v1/chat/completions with model: "bluwing".
  3. Deterministic Choice: Bluwing evaluates the request against your tool index (matching existing bricks or compiling JIT on the fly) and returns:
  4. tool_calls: The exact tools to execute with arguments extracted and grounded in the prompt.
  5. needs_input: A clarifying question because a required argument is missing.
  6. no_tool: Fast deadband response when the request is general conversation or out-of-scope.
  7. Execution: Your backend executes the returned calls with your existing API keys, databases, or microservices.
Note

Bluwing never executes your tools and never handles your sensitive API credentials. It only decides which tools to call and extracts the exact arguments.

#Core Concepts

PieceWhat it is
ToolA standard function definition: name, description, and JSON Schema parameters. Identical to OpenAI and Anthropic tool formats.
Model (bluwing)The unified routing model. Automatically resolves tools against ready bricks or compiles ephemeral bricks on the fly.
Toolset (Brick)A collection of tools compiled into an optimized high-dimensional vector search index.
Dynamic MaskingPassing a subset of tools runs on the parent brick with 0ms compilation delay while masking non-passed tools.
Tokens & MeteringBilled at $0.015 per 1,000,000 input tokens. Output tokens are completely free.

#What You Need to Get Started

  • An Account: Sign up at bluwing.si to receive 5,000,000 Free Tier tokens immediately.
  • Your API Key: Created on the API Keys page. Store it securely in your backend environment variables (BLUWING_API_KEY).
  • Your Tool Definitions: Ordinary JSON Schemas you already use with OpenAI, Anthropic, or LangChain.

The API runs at https://api.bluwing.si. Head to the Quickstart to make your first live call in under five minutes.

Tip

You can test tool resolution live without writing code in the Playground.

Next Quickstart

Something unclear or missing? Write to [email protected] and we'll fix the page.

On this page

  • How It Works in Your Architecture
  • Core Concepts
  • What You Need to Get Started