Braidrun's workflow engine and SDK are now open source
Run the Apache-2.0 workflow engine in your own environment, and connect existing applications to the Braidrun cloud platform with 13 official SDK clients. Both paths are open source on GitHub and connect to the same workflow capabilities.
braidrun-workflow: the open workflow runtime
An engine that runs AI workflows defined in YAML, from the command line or from Kotlin code — the same runtime the cloud platform runs behind the scenes.
The open-source repository contains the complete workflow runtime and command-line tool. It runs the steps defined in your YAML in order — agents, code, tool calls and sub-workflows — and isolates them with Docker subprocesses in production.
- Workflows defined in YAML, runnable from the CLI or Kotlin code
- Agent steps, code steps, parallelism, conditions, iteration, retries, state machines and sub-workflows
- Built-in agent presets for coding, research, writing, data analysis, documents, browser automation and more
- Tool groups for files, shell, Git, HTTP, browser, documents, databases, RAG, email and more
- Delegate to external coding agents through Claude Code and OpenAI Codex subprocesses
- Native subprocesses for local development, Docker subprocesses for production isolation
- Integrates Anthropic, OpenAI, DeepSeek, OpenRouter, Z.ai and NVIDIA models via Koog
- Can run as an MCP server, exposing tool groups to other clients
# Build the command-line tool
./gradlew installDist
# Validate and run the example bundled in the examples directory
./build/install/braidrun-workflow/bin/braidrun-workflow run \
examples/workflows/research-summary.yamlname: research-summary
version: 1.0.0
description: Simple LLM-backed workflow using the researcher preset.
agents:
researcher:
preset: researcher
overrides:
max_iterations: 32
workflow:
- step: summarize
agent: researcher
input: "Explain what Braidrun Workflow is and list three use cases."braidrun-sdk: Connect to Braidrun in a language you know
Thirteen type-safe clients across 12 languages turn the open platform API into familiar methods, models, and error types for your stack.
Use the SDK from an existing application to call the Braidrun cloud platform: start workflows, check execution results, read artifacts, handle human approvals, and integrate webhooks. All clients are generated from the same OpenAPI specification, keeping their interfaces and data models aligned.
- Find workflows and start executions asynchronously
- Read execution status, final results, and artifacts, and cancel active executions
- Review, approve, or reject human approvals, and manage and trigger webhooks
- Typed APIs and data models with support for Bearer, X-API-Key, and permission scopes
- Every language directory includes installation, configuration, endpoint, and model documentation
As a command-line tool, or embedded in your own project
The same parser and executor: run it directly in the terminal, or depend on it as a Kotlin/JVM library.
After building, you get a braidrun-workflow command that validates, dry-runs and executes workflows — and can run a single preset agent without a workflow file. Ideal for scripting and automation pipelines.
braidrun-workflow agent \
--preset researcher \
--prompt "Summarize this repository and list three use cases"- run / validate / dry-run cover execution and checking
- list-presets and list-tools show the built-in capabilities
- mcp-server exposes tool groups as an MCP server
Released versions are distributed through JitPack. One dependency line lets you call WorkflowParser and WorkflowExecutor in your own JVM application — the exact same implementation the CLI uses.
repositories {
mavenCentral()
maven("https://jitpack.io")
}
dependencies {
implementation("com.github.Braidrun:braidrun-workflow:1.0.5")
}- WorkflowParser and WorkflowExecutor are ready to use
- Native subprocesses for development, Docker subprocesses for production
- Requires JDK 21 or newer; Docker isolation mode needs Docker
The open-source edition gives you the engine; the cloud platform gives you the whole workbench
Both sides share the same runtime. Row by row, here's what the open-source edition includes and what the cloud platform adds on top.
| Capability | Open-Source Editionbraidrun-workflow | Cloud Platformbraidrun.com |
|---|---|---|
| Workflow Runtime Engine (YAML) | ✓Deploy and run it yourself | ✓Same engine, hosted for you |
| CLI And Library Integration | ✓CLI plus Kotlin/JVM library | ✓Connect through the official SDKs or REST API |
| Visual Editor (Canvas ⇄ YAML) | —No UI — edit YAML | ✓Canvas and YAML stay in sync both ways |
| Conversational AI assistant for building workflows | —Not provided | ✓Generate and edit workflows by chatting |
| Breakpoint debugging and failed-step reruns | —Debug it yourself | ✓Change values at a breakpoint and resume; completed steps aren't billed again |
| Auto-Resume After Service Restart | —Implement it yourself | ✓Auto-recovers after verifying the definition hash |
| Run History And Monitoring | △Build your own logging | ✓Execution history, cost accounting and alerts |
| Templates & Modules | △A few examples bundled in the repo | ✓240+ templates, 120+ modules and a marketplace |
| Human Approval Gates | △Supported at runtime; notifications are your own | ✓In-app / email / API notifications |
| Schedule / Webhook / API Triggers | △Schedule it yourself | ✓Built-in triggers and chained scheduling |
| Team Collaboration And Permissions | —Not provided | ✓Teams, roles and sharing |
| Sandboxing And Secret Management | ✓Docker isolation, managed by you | ✓Hosted sandbox plus secret encryption |
| Technical Support | —Community only, no official support | ✓Professional Technical Support |
| Self-Hosted Deployment | ✓Self-hosted by definition | ✓Private deployment on the Enterprise plan |
If you'd rather manage less infrastructure and build more workflows
The open-source edition hands you the engine to run yourself; the cloud platform adds the interface, collaboration, hosting and support, so workflows ship faster and are easier to maintain.
When the open-source edition is all you need
We don't think everyone should move to the cloud. In the three cases below, the open-source edition is the right choice on its own.
About the engine, SDK, and cloud platform
Are the open-source edition and the cloud platform the same engine?
Yes. The cloud platform is built on the same open-source runtime, adding visual editing, collaboration, run history, monitoring and hosted execution. YAML you get working in the open-source edition can move to the cloud almost unchanged.
What license does braidrun-workflow use? Can I use it commercially?
It's under the Apache-2.0 license, which permits commercial use and modification as long as you keep the license and copyright notices. See the LICENSE file in the repository.
What is the difference between the SDK and the open-source workflow engine?
The SDK lets your application call the Braidrun cloud platform API and is designed for multilingual system integration. braidrun-workflow is a workflow engine and command-line tool that runs locally or on your own servers.
Can I install the SDK directly from each language's package manager?
The SDK is currently distributed through its GitHub source repository, and availability in language-specific package managers may vary. Choose a client and follow the installation instructions in that directory's README.
Does the open-source edition come with technical support?
The open-source project provides no official technical support and relies on the community and documentation. If you need staffed responses and a service-level commitment, use the cloud platform or the Enterprise plan.
Can I start with the open-source edition and migrate to the cloud platform later?
Yes. Many teams validate a workflow from the command line first, then move it to the cloud platform for the visual interface, collaboration and hosting. Workflows are defined in standard YAML, so migration costs little.
Does using the cloud platform lock me in?
The core engine is open source and workflows are defined in standard YAML. You can export a workflow and run it yourself in the open-source edition at any time — there's no lock-in.