Skip to main content
Axon offers secure, production-ready AI models for both code generation and general-purpose tasks, leveraging State-of-the-Art Deep Reasoning and State Machines

​
Benchmarks

​
Architecture

Axon Models are based on open source models from Qwen 3 family, fine tuned on our proprietary dataset and upgraded with deep reasoning and state machine capabilities.

​
Mixture of Experts (MoE) architecture

MoE is a technique that allows the model to dynamically select the expert model to use for a given input. This architecture enables engineering teams to build more scalable and efficient systems by routing tasks to specialized experts, reducing computational overhead while maintaining high performance across diverse workloads.

​
What makes Axon different?

  • Deep Reasoning: Our SOTA Deep Reasoner generates a detailed reasoning process for your requests, detects what needs to be done and how to do it, ensuring all the context is considered and the best possible solution is provided.
  • State Machines: Our SOTA State Machine uses temporal memories to remember your continued flow of usage on what has accomplished and what needs to be completed next.

​
Deep Reasoner

Our State-of-the-Art Deep Reasoner generates a detailed reasoning process for your requests, detects what needs to be done and how to do it, ensuring all the context is considered and the best possible solution is provided.
  • Multi-sources causal graph traversal for inferencing across heterogeneous data sources, enabling root-cause analysis and counterfactual reasoning.
  • Dynamic symbolic grounding via contextual ontologies to map abstract concepts into actionable knowledge representations in real time.
  • Probabilistic logic synthesis with uncertainty quantification to evaluate solution optimality under incomplete or ambiguous input conditions.
  • Hierarchical attention over structured memory to maintain long-range dependencies during complex, multi-step problem decomposition.
  • Meta-cognitive feedback loops that refine internal heuristics based on outcome validation, improving future reasoning trajectories.
  • Real-time web search integration with federated query optimization across multiple search providers for comprehensive knowledge retrieval.
  • Adaptive web content parsing using semantic-aware scrapers that extract structured data from dynamic web sources while respecting rate limits and ToS.

​
State Machine

Our State-of-the-Art State Machine Engine uses temporal memories to remember your continued flow of usage on what has accomplished and what needs to be completed next.
  • Hierarchical semi-Markov decision processes (HSMDPs) for modeling variable-duration states and adaptive task sequencing.
  • Distributed state persistence with vector-clock reconciliation to ensure consistency across asynchronous, concurrent user sessions.
  • Reinforcement learning-driven transition policies that optimize long-term user goal completion over immediate action rewards.
  • Temporal difference learning over latent state embeddings to predict and pre-fetch likely next states for zero-latency transitions.
  • Context-sensitive state compression using learned subroutines to reduce combinatorial state explosion while preserving semantic fidelity.

​
Model Family

axon_mphl2c

Axon 1

General Purpose Model for high-effort day to day tasks
axon-mini_jjac8i

Axon Mini 1

General Purpose Model for low-effort day to day tasks
axon-code_kfrxe2

Axon Code 1

Code Generation Model for high-effort coding tasks

​
Getting Started

​
Get API Key

Get API Key

Generate a new API key

​
API & SDK Integration

curl --request POST \
  --url https://api.matterai.so/v1/chat/completions \
  --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer MATTER_API_KEY' \
  --data '{
  "model": "{{MODEL_ID}}",
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "user",
      "content": "What is Rust?"
    }
  ],
  "stream": false,
  "max_tokens": 1000,
  "reasoning": {
    "effort": "high",
    "summary": "none"
  },
  "response_format": {
    "type": "text"
  },
  "temperature": 0,
  "top_p": 1
}'

​
Data Privacy

MatterAI never trains on your codebase, all data is temporary and deleted automatically.
⌘I