Portable Scientific Runtime for AI Agents

Scientific infrastructure any agent can use to compute, interpret, and reason over experiments.

Datalox is not an "AI cytometry assistant" and not an "agent framework." It is a programmable execution and data layer for biological experiments. Agents become clients of the runtime.

Deterministic scientific compute
Structured experiment state
Agent-side reasoning
Execution Runtime
Experiment State
Signal Compiler
Any Agent

Core Design Principle

Keep compute, state, meaning, and reasoning as separate layers with explicit boundaries.

Tightly Coupled Assistant Pattern

When compute and reasoning are mixed together, systems become hard to port, inspect, and reproduce.

  • The assistant owns both execution and interpretation.
  • Manual state is hidden in prompts or transient context.
  • Changing agents means rebuilding the stack.

Portable Runtime Pattern

The runtime executes science. The state tracks experiments. Agents reason on top through a stable API.

  • Runtime handles deterministic and reproducible computation.
  • Experiment state preserves gates, groups, conditions, and history.
  • Any agent can plug in and reason without owning execution.
Compute State Meaning Reasoning

What Each Layer Owns

Each concern has one job and one source of truth.

01

Compute

Node-graph execution for transformations, clustering, gating, and stats with deterministic outputs and caching.

02

State

Experiment structure that tracks samples, conditions, manual gates, populations, and full analysis history.

03

Meaning

Biological entities convert measurement space into interpretable objects such as populations, phenotypes, and comparisons.

04

Reasoning

Agents consume structured signals and context to generate hypotheses, explanations, and next-step suggestions.

Our Vision

A portable scientific runtime for biological experiments, designed to be agent-agnostic.

01

Data Layer

Raw measurements only: FCS files, microscopy images, sequencing counts, protein sequences.

02

Scientific Execution Runtime

Operation graph for load, compensate, transform, cluster, gate, and compute statistics.

03

Experiment State + Biological Entities

Track scientist actions, then map numbers into populations, marker profiles, and phenotypes.

04

Signal Compiler

Emit structured evidence like frequency changes, marker shifts, gate shifts, and outlier scores.

Layer 1: Data

Store measurements, no interpretation.

Layer 2: Scientific Runtime

Deterministic compute engine.

Layer 3: Experiment State

Samples, gates, conditions, history.

Layer 4: Biological Entities

Population and phenotype objects.

Layer 5: Signal Compiler

Structured evidence, not conclusions.

Layer 6: Runtime API

Stable tool interface for any agent.

Layer 7: Skills

Reusable analysis workflows.

Layer 8: Agent

Interpretation, hypotheses, suggestions.

Layer 9: UI

Scientist and agent collaborate in the loop.

Full Architecture

From measurement to interpretation, each stage is explicit and replaceable.

Runtime Responsibilities

Deterministic computation, reproducibility, caching, and scalable execution for scientific workflows.

Experiment Representation

An explicit experiment graph containing samples, conditions, gating trees, and derived populations.

Evidence Extraction

Compile raw outputs into reusable signals: PopulationChange, MarkerShift, DistributionShift, GateShift.

Agent-Agnostic Interface

Expose runtime operations through a stable tool surface that any agent can call.

Portable Runtime API

  • Initialize and manage experiment state across samples, conditions, and analysis history.
  • Run scientific workflows for gating, transforms, clustering, and population statistics.
  • Return structured evidence so external agents can interpret results and reason on top.

Skills Layer

  • Reusable workflows such as QC analysis and immune population annotation.
  • Skills can run inside your platform, inside an agent, or as shared modules.
  • Each skill orchestrates context retrieval, signal retrieval, analysis, and interpretation.

Human-Agent Collaboration Loop

  • Scientist action updates experiment state and runtime outputs.
  • Signals are recompiled and surfaced to the reasoning agent.
  • Agent proposes hypotheses and suggestions; scientist accepts or edits.

Why This Structure Enables Portability

The core system is infrastructure. Agents connect through the API, so reasoning can evolve without rewriting scientific execution.

Mental Model

  • Runtime: executes scientific compute.
  • State: tracks the experiment.
  • Agent: interprets signals and context.

Long-Term Direction

  • Start with flow cytometry and structured experiment execution.
  • Expand compute nodes to broader biological modalities over time.
  • Evolve into an AI-native scientific runtime for programmable experiments.
  • Support many agents the way databases support many applications.

Connect With Us

We are building portable scientific infrastructure for agent-assisted biology. If you want to integrate your agent or pilot this in real workflows, reach out.

Company

Complexity LLC

Founded

2025