---
id: engines/timelike
about:
  label: TimeLike
  description: >-
    Steve Coy's simulation framework for decision-making: composable causal
    models of a system, run faster than real time to predict the consequences of
    possible courses of action.
  tags:
    - simulation
    - decision support
    - model-based optimization
    - event-driven
tool:
  tier: 4
  form: simulation framework
  group: Agents and system dynamics
  homepage: 'https://timelike.systems/'
open:
  source: closed
  data: unknown
  api: unknown
  participation: partly
domains:
  - people
provenance:
  origin: agent
  sources:
    - title: TimeLike™ Systems
      url: 'https://timelike.systems/'
    - title: TimeLike Systems / About Us
      url: 'http://www.timelikesystems.com/about.html'
    - title: >-
        Steve Coy's 'Timelike Systems' ; owning the clock. Anselm Hook,
        2026-10-16
      url: 'https://anselm.substack.com/p/steve-coys-timelike-systems-digital'
    - title: Steve Coy - Foresight Institute
      url: 'https://foresight.org/people/steve-coy/'
  created: '2026-10-17'
  reviewed: '2026-10-17'
  reviewed_by: Claude; the four pages above were read on 2026-10-17
  status: draft
---
Steve Coy founded TimeLike Systems in 2011.[^2] The framework descends from tempus, the general-purpose simulation framework Coy led at MZA Associates, whose optics tool WaveTrain became a standard in high-energy laser research.[^2]

The company calls its approach SECHI, simulation-enabled cooperative human intelligence.[^1] Experts encode what they know as causal models, models of what causes what, and these are coupled into a model of a larger system.[^1] The simulation predicts the probable consequences of each course of action open to the user.[^1] Model-based optimisation then narrows them to the course with the best expected outcome.[^1]

The engine is event-driven. Models are blocks with typed ports.[^3] A block does nothing until something it is connected to changes, asks it for a value, or a scheduled event fires.[^3] So each block can carry its own physics and run at its own pace.[^3] Queries carry time, and causality falls out of the coupling.[^3] Components negotiate their fidelity with their neighbours instead of an expert tuning every run.[^3] The aim is a virtual clock that runs faster than the real one: a model fast enough to predict can act as a controller for a place.[^3]

TimeLike contains no machine learning of its own.[^1,3] The company presents it as intelligence augmentation and distinguishes it from generative AI.[^1] A patent application, "Composable Self-Modeling Cyber-Physical Systems", is pending.[^1,3] Earlier tools in the lineage ran satellite-operator training in real time and wrapped a legacy Fortran adaptive-optics code.[^3]

TimeLike is not yet released.[^1] The company is looking for collaborators, strategic partners and early adopters.[^1,4] There is no public download, and pricing is not published.[^1]