Know the best move,
before you make it.
Most enterprise AI executes decisions you've already made. AUTOBUS weighs the alternatives first — simulating each against your goals and constraints — then executes the plan you choose, deterministically.
Supported by
Built by a team from
What AUTOBUS does
Built around the initiative, not the department.
Traditional enterprise systems optimize isolated workflows. LLM agents add flexibility but drift over long, rule-heavy reasoning. AUTOBUS treats the business initiative as the unit of work — an end-to-end objective with its tasks, rules, data, metrics, and actions.
Structured task networks
Initiatives are represented as tasks with explicit preconditions, postconditions, required data, and API-level actions — so dependencies across teams are visible instead of implied.
Alternatives simulated, not guessed
AUTOBUS represents business state, the dynamics between your entities, and the action space actually open to you. Candidate plans are run forward and weighed against your objectives and constraints — before anything touches a live system.
Executable logic, generated
Your entities, relationships, and constraints become logic facts and foundational rules. Agents translate business language into task-specific predicates you can read.
Deterministic execution
A logic engine enforces the constraints and runs the program, orchestrating APIs, models, and domain agents. Same inputs, same outcome — every run.
How it works
From specification to execution, in six moves.
Each step is inspectable. Nothing is hidden behind a model call you can't account for afterwards.
Specification
Your team defines the objective, tasks, data, success metrics, and constraints.
Semantic grounding
Enterprise data is translated into logic facts and foundational rules tied to your business entities.
Simulation
Candidate action sequences are run forward against the modelled business state and its dynamics, then compared on your objectives and constraints. You approve the plan that proceeds.
Logic generation
Agents generate task-specific rules and predicates, along with the data retrieval and actions each task needs.
Execution
The logic engine runs the program deterministically, checking feasibility and constraints before it acts.
Evaluation
Metrics and evaluation rules run continuously, validating outcomes and guiding the next iteration.
Approach
Start simple. Add capability only when the problem demands it.
We'd rather start you where the problem actually is than sell you the largest system on day one.
Get your data to agree with itself
Shared business meaning, trusted datasets, clear ownership and lineage. Often enough on its own — when the real bottleneck is “whose definition of active subscriber do we use?”
Execute known decisions across silos
For when you already know the rules, but running them spans teams, systems, approvals, and audit logs. Turns a known policy into a governed, executable workflow that survives review.
Compare options before committing
Business state, the dynamics between your entities, and the actions actually available to you, represented explicitly — so alternative sequences can be run forward and weighed against goals and constraints, for the cases where the best move genuinely isn't known yet.
Use the simplest layer that solves the problem.
Why it matters
Transformation fails at the seams.
Cross-functional initiatives break down where teams hand off to each other. AUTOBUS is built to make those seams reconfigurable.
Time to market
Update task instructions and policies, regenerate the logic, redeploy — without rebuilding a bespoke pipeline for every change in the business.
Auditability and compliance
Decisions are explicit predicates and rules executed deterministically, so “why did this happen?” has an answer you can hand to a regulator.
Less scarce engineering
Shift effort from hand-coding orchestration to defining semantics, policies, and intent — with agents handling rule construction and integration plumbing.
Research
Built on published research.
AUTOBUS originated from AI research. Each layer of the architecture was worked out and published on peer reviewed scientific journal or conference first.
Toward Data Systems That Are Business Semantic Centric and AI Agents Assisted
Argues that data systems should be organized around what the business means rather than what the tooling makes convenient — curated entities, relationships, and ownership. The substrate every layer above depends on.
Read the paper →Autonomous Business System via Neuro-symbolic AI
The architecture behind the product. LLM agents synthesize task-specific logic programs; a predicate-logic engine enforces constraints and executes them deterministically over business-semantic enterprise data.
Read the paper →Business World Model
Adapts the world-model idea from vision and robotics to an environment that is semantic and organizational rather than physical — representing business state, dynamics, and feasible actions so alternatives can be simulated before committing.
Read the paper →Governance
Powerful, without giving up control.
People stay accountable for business meaning, policy, and high-impact calls. The system accelerates everything underneath that.
Before execution
- Define business semantics — ontology and reference data models
- Set the policies and constraints the engine must enforce
- Curate the tool ecosystem: APIs, models, and domain agents
- Decide which steps require a human approval gate
During execution
- Every decision path is explicit — predicates, rules, and outcomes
- High-impact decisions route to a human before they act
- Missing facts become blockers and escalations, not confident guesses
- Evaluation metrics feed back into the next iteration
FAQ
Questions we hear from evaluation teams.
Is AUTOBUS a workflow engine?
It orchestrates initiatives, but differs from a traditional workflow engine: task logic programs are generated from natural-language instructions plus your enterprise data, then executed deterministically by a logic engine — rather than hand-drawn as a diagram and maintained by hand.
What makes it deterministic if LLMs are involved?
LLMs generate the task predicates and rules and help ground semantics, but they don't execute anything. Execution happens in the logic engine, which enforces constraints and runs the program consistently, producing a transparent decision path you can replay.
What does it mean to simulate a business decision?
AUTOBUS can represent your business explicitly: its current state, the dynamics between entities, and which actions are actually available. An action changes entity attributes directly, and other attributes indirectly through those dynamics — so alternative sequences can be run forward and compared against your stated objectives and constraints before anything touches a live system. It turns “why this action?” into “because simulation predicts it best satisfies the objective you stated, under the constraints you set.”
Do we need a knowledge graph to start?
Not a complete one. AUTOBUS is strongest when data is organized around business entities and relationships with explicit constraints, but most organizations start by linking existing data to a handful of key entities and expanding from there.
How does it connect to our systems?
Through a tool layer of APIs, machine learning models, and domain-specific agents. Tools fetch missing facts — real-time data, for example — and take actions such as updating records, triggering campaigns, or persisting outputs.
When is AUTOBUS the wrong choice?
When your existing systems plus manual coordination already hit the reliability and time-to-market you need at lower cost, or when the generated logic would need so much review that writing the workflow by hand is simply faster. We'd rather tell you that during the pilot than after it.
Get started
Run one initiative end to end.
We're looking for teams with cross-functional, data-rich initiatives where deterministic execution and auditability actually matter. A pilot scopes to one initiative, a small set of entities, and one to three tools.