Repolode
Submit
SynaLinks

SynaLinks/synalinks

Keras-style Python framework for building, training and deploying LM-based apps such as RAG systems, tool-using agents and MCP servers.

View on GitHubWebsite
Stars
464
Forks
47
Open issues
1
Last commit
9 hours ago

What is it?

Synalinks is a neuro-symbolic framework for building language model applications such as RAG pipelines, autonomous agents and multi-agent systems. Programs are composed from modules in a declarative, Keras-like API, so the same structure used for deep learning layers applies to LM workflows.

Programs can be optimized with in-context reinforcement learning using familiar compile, fit, evaluate and predict calls. Prompts, few-shot examples and other trainable variables are tuned per module without changing model weights. Outputs are constrained to JSON data models, and programs can be saved as JSON and deployed as REST APIs with FastAPI or as MCP servers with FastMCP.

The framework routes model calls through LiteLLM, so providers such as OpenAI, Anthropic, Gemini and Ollama can be swapped with a string identifier. It includes an embedded graph database and SQL knowledge base, a sandboxed runtime for agent code execution, and MLflow-based observability callbacks.

Key features

  • 01Keras-style composition of modules for LM programs.
  • 02In-context reinforcement learning that tunes prompts and few-shot examples without changing model weights.
  • 03Constrained JSON outputs defined with data models.
  • 04Model switching across many providers through LiteLLM.
  • 05Deployment as REST APIs with FastAPI or MCP servers with FastMCP.
  • 06Embedded graph database and SQL knowledge base for RAG and memory.

Use cases

  • Building retrieval-augmented generation systems over documents or relational data.
  • Prototyping tool-using agents and multi-agent workflows with structured outputs.
  • Optimizing prompts for a task using a dataset and a reward function.
  • Exposing an LM program as a REST API or MCP server.

In collections

Open-source AI agent frameworks & appsReinforcement learning frameworks & training environmentsFrameworks for building apps and websites

Topics

#llm#framework#rag#agents#keras#reinforcement-learning#neuro-symbolic#prompt-optimization#context-engineering#in-context-learning

FAQ

What is synalinks?
Keras-style Python framework for building, training and deploying LM-based apps such as RAG systems, tool-using agents and MCP servers. Synalinks is a neuro-symbolic framework for building language model applications such as RAG pipelines, autonomous agents and multi-agent systems. Programs are composed from modules in a declarative, Keras-like API, so the same structure used for deep learning layers applies to LM workflows.
Is synalinks free to use? What license does it have?
synalinks is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is synalinks still maintained?
Actively maintained. The last commit to synalinks was on Nov 7, 2026.
What can I use synalinks for?
Building retrieval-augmented generation systems over documents or relational data. Prototyping tool-using agents and multi-agent workflows with structured outputs. Optimizing prompts for a task using a dataset and a reward function. Exposing an LM program as a REST API or MCP server.
Where can I get synalinks?
The source code and installation instructions are on GitHub: https://github.com/SynaLinks/synalinks

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects