Skip to main content
Fireweave (fireweave.ai) is an AI release engineer that takes every coding-agent change to production safely with control points, cohort rollouts, guardrails, and live stability and adoption signals. These are the developer docs: SDKs for Node, Python, Go, Java, Rust, and the browser, plus onboarding for the CLI and coding-agent plugins. Fireweave SDKs evaluate control points and optionally register targets. Your application authenticates with a Fireweave project key and talks to fw-server. The wire protocol and canonical envelopes still use the parameter name flagKey; the product name is control point. v1 exposes exactly those two capabilities. The SDKs do not ship OpenFeature providers, and they do not expose releases, exposures, signals, or capabilities. Guardrails are evaluated by fw-server and the Fireweave console during rollouts; the SDK has no guardrails API. If you are using a coding agent to integrate Fireweave, start with For coding agents.

Onboarding

Account, CLI, coding-agent plugin, then ship your first change behind a control point.

SDK quickstart

Install, initialize with a required mode, evaluate a control point, register a target, and shut down.

SDKs

Seven language bindings share one surface: a mode-selected adapter, a runtime, and a client with controlPoints (cased per language) plus registerTarget.

Node.js, Bun, and Deno

@fireweaveai/server-sdk 2.3.0 — initFireweave, controlPoints, registerTarget. Node >=20.20.

Python

fireweave 2.3.0 — init_fireweave, control_points, register_target. Python >=3.10.

Go

…/sdks/go/v2 v2.3.0 — Init, ControlPoints(), RegisterTarget. The /v2 suffix is required.

Java

ai.fireweave:fireweave-sdk 2.3.0 — Fireweave.init, controlPoints(), registerTarget(). Java 11+.

Browser

@fireweaveai/web-sdk 2.3.0 — synchronous reads, registerTarget / identify, fw_public_ keys.

Rust

crates.io fireweave 2.3.0 — init_fireweave, control_points, register_target. Rust 1.75+.

Compatibility

Seven-language surface, 65×7 conformance aggregate, and the capability matrix.

Package index

Registry names, versions, and what is actually installable.

What you can do

  • Evaluate boolean, string, number, and object decisions through nine methods. A read returns your default on failure and does not throw.
  • Register targets once at login (registerTarget / language equivalent) so targeting rules can match durable properties.
  • Choose a mode at init: remote (fw-server) or local (seeded in-process map). mode is required and never inferred.
The product noun is control point. flagKey remains the key name on the wire, in canonical envelopes, in conformance fixtures, and in capabilities.features.flags. The onboarding guide covers the product path — plugin skills and gradual rollouts — while these pages are the SDK reference.

Architecture and concepts

How it works

Evaluate and register — the runtime path from your app to fw-server.

Architecture

Client → Runtime → Adapter, lifecycle states, and the fw-server paths.

Core concepts

Control points, targeting, adapters, and decisions.

Migrate to v1

Package rename, required mode, and the retired OpenFeature / extension surface.

Production, troubleshooting, reference

Configuration and auth

Explicit apiUrl / apiKey (or language equivalent). The SDK reads no environment variables.

Lifecycle

Initialize, READY / STALE, shutdown. Reads after close return defaults.

Errors

Fifteen error kinds and how reads degrade to your default.

Troubleshooting

Defaults that look like “off,” missing targeting keys, and boot configuration errors.

HTTP API

POST /v1/flags/evaluate, POST /v1/capture, POST /v1/targets/register.

CLI

Command-line tools for Fireweave projects.
Start with Onboarding to ship a first Fireweave rollout, or the Quickstart for a first evaluate in one sitting.
Last modified on September 12, 2026