System Components

This page explains how our core components work together to create the near-perfect point-in-time context essential for entropy stratification.

Key idea – entropy stratification We match the precision of reasoning to the uncertainty of the situation. Components cooperate so agents know when lightweight heuristics suffice and when to escalate to high-fidelity reasoning.

Component Map

Component	Primary role	Core question it answers
Agent Core	Defines stable professional identity and judgment standards	Who is acting, and what guarantees do they uphold?
Context Graphs	Encode the navigable problem topology	Where are we in the workflow, and which paths are legal?
Functional Memory	Maintains outcome-sufficient history	What should we remember right now, and at what resolution?
Dynamic Behaviors	Provide adaptive modifiers	How should we adjust execution when reality deviates from the base plan?
Actions	Orchestrate tools and workflows	What should we execute right now, and how do we do it reliably?
Evaluations	Measure performance and govern change	Did it work, and what evidence justifies the next update?

1

Agent Core (foundation)

Start here to understand the stable identity and expertise that anchors the system.

2

Context Graphs (structure)

Learn how the problem space is defined and organized.

3

Functional Memory (persistence)

Understand how context is maintained over time.

4

Dynamic Behaviors (adaptation)

Discover how the system adapts with real-time flexibility.

5

Actions (execution)

See how the system executes work through controlled tooling.

6

Evaluations (measurement)

See how the system measures outcomes and governs change.

The components integrate to form the unified context that enables intelligent decision-making.

---

Agent

Agent Core

The agent core sets a durable professional identity—scope of practice, escalation posture, communication style—that interprets every measurement. Identity stays latent until the context graph activates it, which is why the same agent can offer different behaviors in different states. Learn more in Agent Core.

Context Graphs

Context graphs supply the navigational map for a service: the intents, legal transitions, and guardrails that keep a workflow safe. They are best understood as the topology that the agent walks—triage queue → risk assessment → escalation—while the dimensional blueprint (described later) tells us what to measure about the patient at each point. Graphs can be exploratory or highly scripted, but they only become operative once they combine with identity, memory, behaviors, and actions. We break the conceptual, structural, and operational layers down in Context Graphs.

Dynamic Behaviors

Dynamic behaviors adapt the problem space in real time. They can adjust optimization targets, widen or narrow entry predicates, request deeper reflection, or expose tools when the measured state leaves a validated band. In short, they let a single context graph handle everything from routine flows to rare edge cases. See Dynamic Behaviors.

Functional Memory

Functional Memory operationalizes the dimensional blueprint for the object of care (for example, a patient). It ensures that every feature the blueprint names—vital signs, medication timelines, staffing signals—is captured, aligned, and ready for live reasoning. The system centers around user models derived from custom dimensional frameworks that organizations design to interpret raw information through clinical interactions.

Unlike traditional approaches that treat all information equally, our dimensional framework organizes memory according to functional importance, determining what information requires outcome-sufficient preservation (maintaining sufficient statistics—compressed representations preserving all information relevant to outcomes), how contextual relationships should be maintained over time, and when information should be recontextualized based on new understanding.

The memory system operates through a hierarchical architecture (L0→L1→L2→L3) that compresses thousands of observations into 10-50 functional dimensions driving outcomes, preserving what matters while discarding noise. This functional alignment ensures agents have all the context they need for optimal entropy assessment and decision-making without constant information retrieval.

Memory doesn't operate alone—it combines with professional identity (interpretation priors), context graphs (problem structure), and constraints to form the unified context that enables decisions. The hierarchical compression maintains sufficient statistics at each layer while preserving the ability to replay raw traces when dimensional blueprints evolve.

For more details, see Functional Memory.

Actions

Amigo Actions represent the execution layer that transforms our orchestration framework into real-world outcomes through custom programs running in isolated execution environments. Unlike traditional tool calling, Actions can orchestrate entire workflows—authenticating with external systems, processing data through multiple steps, handling errors and retries, and coordinating between different services. The LLM provides contextual reasoning about what needs to happen, while Actions handle the deterministic execution.

Context-aware integration allows sophisticated Action composition and orchestration. Different states in a context graph expose different capabilities—when a clinical agent focuses on emergency triage, it has access to vital sign analyzers, but when transitioning to treatment planning, different Actions become available like drug interaction checkers and care protocol analyzers. Dynamic behaviors can modify the available Action landscape in real-time based on conversational context, creating a fluid, adaptive tool environment where capabilities evolve based on specific problem contexts.

For more details, see Actions.

Platform

Evaluations

Evaluations define what “good” looks like for each problem neighbourhood. They run persona-driven and adversarial simulations, score multi-objective outcomes, and track admissibility margins so we know how close a configuration is to breaching safety or value constraints. Results feed directly into optimisation decisions. See Evaluations.

Pattern Discovery

Pattern discovery fine-tunes the behaviors that evaluations approve. It concentrates on high-leverage adjustments—thresholds for switching reasoning gears, timing for memory expansion, balance between autonomy and structure—while routine control stays with the baseline orchestration. Details live in Pattern Discovery and Optimization.

How the components work together

Session level. Interactions follow quantum patterns such as [A] -> [A] (direct response) or [A] -> [D] -> [R] -> [A] (decision and reflection before speaking). Within those patterns, identity interprets measurements, context graphs decide which states are legal, memory supplies outcome-sufficient context, behaviors adjust the plan, and actions execute the deterministic work.

System level. Evaluations surface where the composition succeeds or fails. Pattern discovery proposes measured improvements. Agent Forge promotes approved changes while keeping a rollback path.

Agent Forge

Agent Forge is the control plane for recursive improvement. It versions configurations as code, lets coding agents explore alternatives inside safe sandboxes, and requires human approval before production rollout. The result is a system that can evolve quickly without giving up observability or governance.