# InferTheta

> InferTheta is a decision infrastructure platform that uses probabilistic graphs and Bayesian inference to govern AI agents and autonomous hardware.

Design probabilistic decision graphs visually, then deploy them (live API endpoints, Python export, and more) for AI agents and autonomous hardware. Give agents and machines structured causal reasoning, auditable decision boundaries, and hallucination-free action selection under heavy uncertainty.

## Agent API (machine entry)

Remote agents should use the production Agent API — not this marketing site SPA.

- Base: https://infertheta-agent-engine-production.up.railway.app/
- Agent index: https://infertheta-agent-engine-production.up.railway.app/llms.txt
- OpenAPI: https://infertheta-agent-engine-production.up.railway.app/openapi.json
- Pricing: https://infertheta-agent-engine-production.up.railway.app/v1/pricing
- Free sandbox: `POST /v1/sandbox` → `apiKey` + small credits (no Auth0, no card)
- Paid credits: `POST /v1/checkout` `{ "pack": "starter" }` → human opens `checkoutUrl` → `GET /v1/checkout/{sessionId}` returns `apiKey` once
- Auth: `Authorization: Bearer <apiKey>` or `X-API-Key: <apiKey>`
- Workflow: create/edit networks → evidence → inference → evaluate-decision → explain
- Humans: [https://infertheta.com/for-agents](https://infertheta.com/for-agents) — why fund the Agent API. Machines: https://infertheta-agent-engine-production.up.railway.app/llms.txt

## What InferTheta is (and is not)

- **Is:** decision infrastructure — probabilistic graphs / Bayesian networks that govern what an agent or machine may do given what it knows.
- **Is not:** a general chat LLM, a black-box classifier, or a replacement for your existing agent/robot stack. Callers query the graph with observations and receive bounded, explainable actions.

## Our thesis

Adoption of AI agents in critical sectors like defense, finance, and cybersecurity faces a hard ceiling. High-stakes operations cannot tolerate unconstrained systems, and LLM decisions remain difficult to explain due to persistent hallucination risks.

The same barrier blocks physical systems like drones, robotics, and autonomous vehicles. Until engineers can deterministically govern, audit, and explain agentic reasoning under uncertainty, true autonomy in critical operations will remain out of reach.

## Our solution

InferTheta forces AI agents and autonomous hardware to execute critical decisions through structured probabilistic graphs. When an agent encounters a volatile situation, it maps its active observations (and known unknowns) into the graph. The engine calculates exact posterior probabilities and outputs a mathematically grounded course of action.

### Modelling Engine

Design probabilistic graphs visually with no code: variables, causal structure, CPDs, observation docs, and agent decision rules. Learn probabilities from data, then run exact inference in-browser or on the edge.

### Governed Agent Decisions

Export graphs that AI agents invoke as native tools, or call the live Agent HTTP API (sandbox or pay-per-call credits). Incoming observations update posterior probabilities in real time, while custom threshold rules map beliefs directly to bounded actions. Every agent decision remains fully explainable, auditable, and constrained.

### Runtime for Machines

Ship models directly as standardized JSON or idiomatic Python code. Embedded controllers, field robotics, and autonomous hardware follow a strict observe → infer → act loop locally, executing exact probabilistic reasoning without relying on opaque neural weights.

## Who it serves

From LLM agents handling complex workflows to field robotics navigating unstructured environments, InferTheta provides a shared reasoning engine for any autonomous system making high-stakes choices.

### High-Stakes Software (AI Agents)

- **Defense & National Security:** Unmanned systems, threat assessment, SIGINT/sensor fusion, tactical decision support.
- **Financial Services & FinTech:** Automated trading compliance, high-frequency fraud detection, credit risk gating, anti-money laundering (AML).
- **Cybersecurity & SecOps:** Autonomous incident response, threat triage, active defense policies where false positives break infrastructure.

### Physical & Embedded Systems (Hardware & Edge)

- **Robotics & Autonomous Hardware:** Field robotics, warehouse AGVs, inspection drones, space & subsea exploration.
- **Manufacturing & Industry 4.0:** Predictive maintenance gating, automated quality assurance, safety-critical shutdown triggers.
- **Supply Chain & Logistics:** Autonomous fleet routing, dynamic risk mitigation, port and yard automation under weather/delay uncertainty.
- **Energy & Critical Infrastructure:** Grid load management, nuclear/industrial plant safety automation, pipeline monitoring.

## Advantages

- **Small Data, Full Power:** Agents and robots rarely get clean big data. Probabilistic graphs stay reliable when observations are scarce, noisy, or expensive.
- **No Black Box, Full Audit Trail:** Every inference path is traceable, from priors to posteriors to decision rules, so operators can explain why a machine acted.
- **Total Variable Visibility:** The graph exposes dependencies directly. Agents and humans see which beliefs drove an action, not just a score.
- **Encode Expert Intuition:** Domain expertise becomes priors, observation docs, and constraints, then updates with evidence instead of vanishing into weights.
- **Near-Zero Compute Cost:** Exact discrete inference runs on modest companion compute, practical for edge robots and autonomous hardware.
- **Causal, Not Just Correlative:** Probabilistic graphs model how interventions change beliefs, so callers can evaluate what to do, not only what correlates.

## FAQ

### What is InferTheta?

InferTheta is a decision infrastructure platform that uses probabilistic graphs and Bayesian inference to govern AI agents and autonomous hardware.

### How is InferTheta different from an LLM agent?

LLM agents generate fluent text but can hallucinate and are hard to audit. InferTheta sits as a decision layer: observations update a probabilistic graph, and actions are chosen from exact posteriors and explicit decision rules you can inspect.

### Who is InferTheta for?

Teams building AI agents or autonomous hardware in high-stakes domains—defense, cybersecurity, finance, robotics, manufacturing, supply chain, and critical infrastructure—that need governed, explainable decisions under uncertainty.

### Do I need to write code to use InferTheta?

Humans can design graphs visually in the modelling engine. Agents should use the production HTTP Agent API (sandbox key or paid credits)—no Auth0. You can also export standardized JSON or idiomatic Python for edge / embedded runtimes.

### How do AI agents call InferTheta?

Start at https://infertheta-agent-engine-production.up.railway.app/llms.txt or https://infertheta-agent-engine-production.up.railway.app/openapi.json. POST /v1/sandbox for a free API key, or POST /v1/checkout for paid credits (a human completes Stripe Checkout). Then create networks, set evidence, run inference, and request explanations with Authorization: Bearer <apiKey>.

### Does InferTheta replace my existing agents or robots?

No. Agents and machines call InferTheta graphs as a reasoning / governance tool: observe → infer → act, with bounded and auditable outputs.

## Contact

Want agents, robots, or autonomous hardware that call a controllable and mathematically verifiable decision framework? Contact: [jlosella@infertheta.com](mailto:jlosella@infertheta.com) or https://infertheta.com/#contact-form