FIT Framework: Full-stack AI Development Framework for Java
A multi-language function engine (FIT), a streaming orchestration engine (WaterFlow), and a LangChain alternative for Java (FEL), with native/Spring dual-mode execution, hot-plugging plugins, and smart cluster deployment.
Evidence shows: security policy (SECURITY.md) and SSRF protection tests exist, but no explicit implementation of least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, or source attribution. Deductions: these aspects lack concrete evidence.
Evidence shows: README and documentation structure are consistent, but dependency availability is unverified and failure messages are not explicit. Deductions: dependency availability and failure messages lack concrete evidence.
Evidence shows: clearly targets Java developers, provides multiple scenarios (FIT, WaterFlow, FEL), but trigger precision and boundary descriptions are not detailed. Deductions: trigger precision and boundary descriptions are not detailed.
Evidence shows: README, installation instructions, examples, and license exist, but known limitations, version changelog, and maintenance responsibility are missing. Deductions: known limitations, version changelog, and maintenance responsibility are missing.
Evidence shows: output usability (code examples) and marginal value (filling Java gap) are reflected, but cost-benefit is not quantified. Deductions: cost-benefit is not quantified.
Evidence shows: tests (SSRF test) and documentation exist, but claim traceability, cross-source corroboration, and fact-inference separation are insufficient. Deductions: claim traceability, cross-source corroboration, and fact-inference separation are insufficient.
- Unverified publisher identity, handle with caution.
- Dependency security is not explicit, recommend checking for dependency vulnerabilities.
- Lack of user confirmation mechanism, may automatically execute external actions.
What does this agent do, and when should you use it?
FIT Framework is an open-source, MIT-licensed, enterprise-grade AI development framework for Java, currently at v3.7.0-SNAPSHOT. It consists of three parts: FIT Core (a multi-language function computing base supporting hot-pluggable plugins in Java/Python/C++), WaterFlow Engine (a streaming orchestration engine with both graphical and declarative API), and FEL (the FIT Expression for LLM, providing LangChain-like primitives for Java developers). The framework runs in both native and Spring modes, and its intelligent clustering deployment lets developers switch from monolithic to distributed without code changes, with runtime auto-routing to local calls or RPC. FIT reduces boilerplate through sensible conventions, automatically choosing communication protocols (HTTP/gRPC/shared memory), auto-injecting plugin dependencies, and providing built-in service discovery and circuit breaking. The project includes comprehensive documentation, quick-start guides, an AI coding sandbox, and build scripts, enabling developers to build LLM-powered business applications efficiently.
FIT Framework provides a complete toolchain: FIT Core handles multi-language function computing with hot-pluggable plugins and smart clustering deployment, launched via ./build/bin/fit start; WaterFlow Engine offers declarative syntax and graphical orchestration for flows ranging from microflows to long cross-system transactions; FEL provides standard primitives like AiFlows and AiProcessFlow to wrap LLMs, knowledge bases, and tools, supporting typical AI scenarios such as retrieval-augmented generation and agents, e.g., building a retrieval flow with AiFlows.create().runnableParallel(...).retrieve(...).synthesize(...). The framework also includes an AI coding sandbox (based on Colima + Docker + Git Worktree) to run AI coding tools in isolated containers. Build commands include mvn clean install and ./build.sh, with the latter also compiling the Sandbox CLI.
- Java developers who want to build LLM applications without switching stacks, using FEL primitives for RAG or agent flows.
- Enterprises needing to integrate LLM capabilities with existing business systems, leveraging FIT Core's multi-language function engine and hot-pluggable plugins for modular extension.
- Developers orchestrating complex business processes with WaterFlow's declarative API or graphical interface, from simple to cross-system flows.
- Teams that need to switch flexibly between monolithic and microservices architectures, using FIT's smart clustering deployment without code changes.
- Developers wanting to use AI coding tools (e.g., Claude Code, Codex) in isolated environments, using the
sandboxcommand to create separate workspaces.
What are this agent's strengths and limitations?
- First full-stack AI framework for the Java ecosystem, filling the gap and avoiding stack switching.
- Multi-language function engine (FIT Core) supports hot-pluggable plugins in Java/Python/C++, offering flexible extensibility.
- Smart clustering deployment seamlessly switches between monolithic and distributed, reducing operations overhead and increasing architectural elasticity.
- WaterFlow engine provides both graphical and declarative API, simplifying flow design.
- FEL offers a LangChain alternative for Java, with standard primitives wrapping LLMs, knowledge bases, and tools.
- Built-in service discovery and circuit breaking, plus Spring compatibility, facilitates enterprise integration.
- The project is at SNAPSHOT version, so APIs may be unstable; caution is advised for production use.
- Documentation is primarily in Chinese, with limited English resources, posing a barrier for international developers.
- Setup requires JDK, Maven, and environment variable configuration, which is somewhat involved.
- As a relatively new framework, the community ecosystem and third-party integrations may be less mature.
- Python/C++ support may rely on plugin development, requiring extra effort beyond Java.
How do you install or deploy this agent?
To install FIT Framework, set up the development environment: JDK 17 or 21, Maven 3.8.8+, and IntelliJ IDEA recommended. Clone the repository and run mvn clean install in the root to build the Java framework, or ./build.sh for a full build (default skips tests; use --with-test to include them). The build outputs to build/, including the fit startup script and sandbox AI coding sandbox command. Then set the system environment variable FIT to point to the FIT core framework directory, and create a new plugin directory where you can run ./build/bin/fit start to start the service.
How do you use this agent?
To use FIT Framework, follow the quick-start guides: for FIT function framework, refer to framework/fit/java/README.md to build a basic web app; for WaterFlow engine, refer to framework/waterflow/java/waterflow-core/README.md to learn the declarative syntax and output 'hello world!'; for FEL, refer to docs/framework/fel/java/quick-start-guide/01. Model.md to build an end-to-end LLM application. For example, in Java, create a retrieval flow with FEL: AiProcessFlow<Tip, Content> retrieveFlow = AiFlows.<Tip>create()...retrieve(...).synthesize(...). Ensure you have a plugin directory and environment variable configured before running fit start.
How does this agent compare with similar options?
FIT Framework's README explicitly states that FEL is a LangChain alternative for the Java ecosystem, thus comparing with LangChain from the Python world. Compared to LangChain, FEL is more tailored to Java engineering practices, offering high performance and engineering advantages, but its ecosystem maturity may be lower.