tardygrada

[](https://github.com/fabio-rovai/tardygrada/actions/workflows/ci.yml)

<p align="center"> </p>

LangChain is 50,000 lines of Python. CrewAI is 153,000. LlamaIndex is 237,000. Tardygrada replaces them in 15-53 lines.

tardy terraform /path/to/any-agent-framework
# CrewAI:     153,245 lines -> 53 instructions
# LlamaIndex: 237,414 lines -> 15 instructions
# LangGraph:  101,662 lines -> 39 instructions
# MetaGPT:     89,734 lines -> 11 instructions

Tardygrada is a programming language where every value is a living agent. Programs compile to MCP servers. Every output is cryptographically verified before it can become a Fact. 212KB binary. Zero dependencies. Pure C11.

Every agent framework does the same thing: call an LLM, pass the result to the next LLM, hope it's correct. None of them verify the output. None of them prove the agent did the work. None of them ground claims against reality.

Tardygrada does. An agent says "Doctor Who was created at BBC Television Centre" and the system decomposes it into triples, grounds it against an ontology, runs 3 independent verification passes with Byzantine majority vote, and only then freezes it as an immutable, cryptographically signed Fact.

If the system doesn't have the knowledge to verify a claim, it says "I don't know" instead of making something up. 0% false acceptance rate without an ontology. 100% correct acceptance with one.

# Build (takes < 1 second)
git clone https://github.com/fabio-rovai/tardygrada && cd tardygrada && make

# Verify a claim
tardy run "Doctor Who was created at BBC Television Centre in 1963"

# Convert any agent framework to Tardygrada
tardy terraform /path/to/crewai

# Run a .tardy program as an MCP server
tardy serve examples/medical.tardy

agent MedicalAdvi