Imandra - AI reasoning with mathematical proof correctness

Introduction to Imandra: The Neurosymbolic AI Platform for mission-critical reasoning

The software industry faces a fundamental challenge: Large Language Models (LLMs) excel at generating code but struggle with proving its correctness. Recent benchmark studies reveal that pure LLM reasoning exhibits a 41-47 percentage point accuracy gap when tasked with software logic analysis, leaving enterprises with code that appears correct but lacks verifiable guarantees. This limitation becomes critical in financial infrastructure, autonomous systems, and defense applications where failure carries regulatory, safety, or existential consequences.

Imandra emerges as the industry's first neurosymbolic AI platform that combines formal mathematical verification with Large Language Models. Unlike traditional AI coding assistants that rely solely on statistical pattern matching, Imandra delivers mathematically provable correctness guarantees for AI-generated outputs. The platform's core differentiation lies in its ability to provide mathematical proofs rather than probabilistic assessments—a fundamental shift for industries where "probably correct" fails regulatory scrutiny.

The company's product portfolio addresses the full spectrum of formal reasoning needs:

• CodeLogician™: AI coding assistant enhancer that transforms source code into precise mathematical logic
• ImandraX: Industrial-grade automated reasoning engine powering verification workloads
• Imandra Universe: Cloud platform delivering Reasoning as a Service™
• Imandra Markets: Specialized financial products for algorithmic trading validation

Imandra's customer roster demonstrates broad trust across safety-critical sectors. Government and defense agencies including DARPA, US Navy, and Sabel Systems rely on the platform for mission-critical verification. Financial institutions such as Goldman Sachs, Citi, KPMG, and Euronext utilize Imandra for algorithmic trading validation and regulatory compliance. Academic partnerships with University of Edinburgh, Cambridge University, and Imperial College London ensure continuous research advancement in automated reasoning.

---

Core Capabilities: How Imandra Transforms AI Reasoning

Imandra's technology suite delivers five interconnected capabilities that address the fundamental limitations of pure machine learning approaches to code reasoning.

CodeLogician: AI Coding Assistant Logic Enhancement

CodeLogician represents a paradigm shift in AI-assisted programming. The tool transforms source code into an exact mathematical model—a MetaModel that preserves functional equivalence with the original implementation. This transformation enables developers to interrogate code behavior with mathematical precision, generate test suites with quantified coverage metrics, and plan source code modifications with verified correctness guarantees.

The performance impact is substantial: CodeLogician bridges the 41-47 percentage point accuracy gap that pure LLMs exhibit on software logic reasoning tasks. This enhancement operates through logical reasoning rather than pattern matching, enabling accurate edge case detection, decision boundary understanding, and bug discovery that training data alone cannot support.

Formal Verification: Mathematical Proof of Correctness

Imandra's formal verification capability uses the ImandraX engine to prove mathematical correctness of algorithms and systems. Unlike traditional testing—which can only demonstrate the presence of bugs, never their absence—formal verification provides mathematical guarantees that properties hold for all possible inputs.

The verification process is fully automated. ImandraX abstracts away the specialized knowledge traditionally required for formal methods, enabling ordinary software engineers to leverage mathematical verification without PhD-level expertise in theorem proving.

Test Case Generation: Automated Edge Case Coverage

Generating comprehensive test suites traditionally requires extensive domain expertise and manual effort. Imandra automatically derives test cases from formal models, producing test suites that cover edge cases and critical scenarios humans frequently overlook. This automated approach ensures回归测试 coverage that would otherwise require weeks of manual test design.

Region Decomposition: State Space Analysis

Inspired by Cylindrical Algebraic Decomposition, Imandra's region decomposition algorithm partitions infinite input domains into finite collections of behaviorally invariant regions. This capability enables complete system behavior mapping for reinforcement learning optimization, reducing training data requirements by identifying regions where system behavior remains consistent.

Counterexample Synthesis: Automated Failure Analysis

When verification hypotheses fail, Imandra generates concrete counterexamples that demonstrate precisely how and why properties break. The reasoning engine is counterexample-complete, supporting recursive functions and nonlinear arithmetic—capabilities that distinguish Imandra from first-order SMT solvers with limited expressiveness.

---

Technical Architecture and Core Engine

Imandra's technical foundation rests on ImandraX, an industrial-grade automated reasoning engine purpose-built for verification workloads in financial, autonomous, and safety-critical systems. The engine implements advanced decision procedures for nonlinear arithmetic, supporting both bounded and unbounded verification scenarios.

The ImandraX Engine

At the heart of Imandra lies the ImandraX reasoning engine—a system developed through contributions to Z3 SMT solver, MetiTarski, and proprietary nonlinear proof procedures. The engine provides several fundamental capabilities:

• Formal Verification: Proves properties about programs using mathematical induction and automated theorem proving
• Counterexample Synthesis: Generates concrete failing inputs when properties cannot be proven, even with recursive functions and nonlinear arithmetic
• Constraint Solving: Solves constraint satisfaction problems involving recursive and higher-order functions
• Region Decomposition: Applies cylindrical algebraic decomposition techniques to partition state spaces

The engine operates on IML (Imandra Modeling Language), a carefully designed subset of OCaml that combines functional programming expressiveness with formal verification tractability. OCaml's type system and mathematical foundations make it ideal for verification—variables cannot be accidentally undefined, and computational behavior matches mathematical semantics precisely.

Interface Layer

Imandra provides multiple integration pathways to fit existing developer workflows:

• VS Code Extension: Full IDE integration for OCaml/IML development with real-time verification feedback
• Python Library: PyPI-hosted library enabling Imandra reasoning capabilities within Python ecosystems
• MCP Server: Model Context Protocol server for integration with AI assistants and custom tooling
• Imandra Universe: Cloud platform providing Reasoning as a Service without infrastructure management

Academic Foundation

Imandra's technology builds on decades of academic research in automated reasoning. The company's co-founder Dr. Grant Passmore (PhD, University of Edinburgh) made significant contributions to Z3 and MetiTarski during his doctoral work on nonlinear arithmetic decision procedures. The company's research continues through active publication, including the recent arXiv paper 2601.11840 presenting benchmark results that demonstrate CodeLogician's effectiveness in bridging LLM reasoning gaps.

---

Ecosystem and Integration

Imandra operates within a broader ecosystem of developer tools, research institutions, and industry consortia that collectively advance formal reasoning adoption.

Developer Tooling Integration

The platform integrates seamlessly with established development environments:

• VS Code: The ImandraX extension provides real-time verification feedback, inline type information, and proof exploration directly within the IDE
• Python Ecosystem: The PyPI-hosted library enables Python developers to leverage Imandra reasoning within existing ML pipelines, trading systems, or automation frameworks
• MCP Server: The Model Context Protocol server allows integration with AI coding assistants beyond CodeLogician, extending formal verification benefits to broader toolchains

Academic Research Partnerships

Imandra maintains active research collaborations with leading institutions:

• University of Edinburgh: Primary research partner, leveraging the Edinburgh Centre for Robotics for autonomous systems verification
• University of Cambridge: Joint research in formal methods and verification techniques
• Imperial College London: Collaboration on probabilistic reasoning and ML verification
• Heriot-Watt University, King's College London, Southampton University: Additional UK academic partnerships
• Budapest University of Technology and Economics: Eastern European research collaboration
• SRI International: US-based research partnership

Industry Consortia and Standards

Imandra participates in industry organizations shaping verification standards:

• Digital Twin Consortium: Member influencing standards for digital twin verification
• ROS Industrial: Alliance member supporting robot operating system verification capabilities

Representative Customer Implementations

Organizations across safety-critical sectors deploy Imandra for various verification challenges:

• Goldman Sachs and Citi: Financial algorithm verification and regulatory compliance
• Euronext and TMX: Exchange trading system validation
• Broadridge: Financial technology infrastructure verification
• DARPA and US Navy: Government defense project verification
• Edinburgh Centre for Robotics: Autonomous vehicle controller verification

---

Pricing Plans

Imandra offers tiered pricing to support users from individual evaluation through enterprise deployment. All plans provide access to the full ImandraX engine capability; differentiation comes through monthly credit allocations and support levels.

Plan Details:

The Free plan provides 100 credits monthly with no credit card required—ideal for exploring Imandra's capabilities and completing the learning curve. The Essential plan at $25/month delivers 1,000 credits with modest overage charges, suitable for individual developers actively building verification workflows. The Growth plan offers 10,000 credits at $200/month with reduced overage rates ($0.018/credit), designed for teams running production verification workloads. Enterprise deployments receive custom pricing with dedicated support channels, infrastructure customization, and SLA guarantees.

Importantly, Free and Essential plans require no credit card information, lowering barriers to evaluation.

---

Frequently Asked Questions