CodeLogician
CodeLogician is the neurosymbolic agentic governance framework for AI-powered coding.
It helps your coding agent think logically about the code it's producing and test cases it's generating.
The fundamental flaw that all LLM-powered assistants have is the reasoning they're capable of is based on statistics,
while you need rigorous logic-based automated reasoning.
- Generated code is based on explainable logic, not pure statistics
- Generated test cases are generated come with quantitative coverage metrics
- Generated code is consistent with the best security practices leveraging formal verification
To run CodeLogician, please obtain an Imandra Universe API key available (there's a free starting plan)
at Imandra Universe and make sure it's available in your environment as IMANDRA_UNI_KEY.
Three typical workflows:
- DIY mode - this is where your agent (e.g. Grok) uses the CLI to:
- Learn how to use IML/ImandraX via
doc command (e.g. codelogician doc --help)
- Synthesizes IML code and uses the
eval command to evaluate it
- If there're errors, use
codelogician doc view errors command to study how to correct the errors and re-evalute the code
- Agent/multi-agent mode - CodeLogician IML Agent is a Langgraph-based agent for automatically formalizing source code into Imandra Modeling Language (IML).
- With
agent command you can formalize a single source code file (e.g. codelogician agent PATH_TO_FILE)
- With
multiagent command you can formalize a whole directory (e.g. codelogician agent PATH_TO_DIR)
- This is a "live" and interactive version of the
multiagent command
- It monitors the filesystem and fire's off formalization tasks on source code updates as necessary
- You can start the server and connect to it with the TUI (we recommend separate terminal screens)
Learn more at CodeLogician!
To get started,
codelogician --help