Augmented LLM

concept Apr 12, 2026

ai-agentsllmretrievaltoolsmemorybuilding-block

The foundational building block of agentic systems: a language model enhanced with retrieval, tools, and memory. Rather than using a bare LLM, every agentic pattern composes one or more augmented LLMs that can actively search for information, invoke external capabilities, and retain state across interactions.

Three augmentations

Augmentation	What it does	Model behavior
Retrieval	Access to external knowledge sources	Model generates its own search queries
Tools	Interaction with APIs and services	Model selects and invokes appropriate tools
Memory	Persistent state across calls	Model determines what information to retain

Current models perform all three actively — they are not passive recipients of injected context, but participants that direct the use of their augmentations.

Implementation priorities

Two aspects matter most:

Tailor capabilities to the use case. Generic retrieval or tool sets underperform purpose-built ones.
Provide a well-documented interface. The model’s effectiveness scales with how clearly its capabilities are described. This is the foundation of the Agent-Computer Interface concept.

Model Context Protocol

MCP is one approach to implementing augmentations — a standard protocol that lets developers integrate with third-party tools through a simple client. The key value is ecosystem interoperability: one integration works across all MCP-compatible tools.

Role in the pattern hierarchy

The augmented LLM sits at the base. All five agentic workflow patterns compose augmented LLMs:

Prompt chaining: sequential augmented LLM calls
Routing: classifier augmented LLM + specialized augmented LLMs
Parallelization: multiple augmented LLMs running concurrently
Orchestrator-workers: orchestrator augmented LLM delegating to worker augmented LLMs
Evaluator-optimizer: generator + evaluator augmented LLMs in a loop

Connections

The “brain” in Brain-Hands Decoupling is an augmented LLM — Claude plus harness loop, with sandboxes as the tools
Hermes Agent is a concrete implementation: Claude + 40 tools + FTS5 memory + skills = one augmented LLM in a chat loop
Context Window Compression addresses the memory augmentation — what happens when the model’s retained state exceeds capacity
The concept connects to the LLM wiki pattern in LLM Knowledge Bases where the retrieval augmentation is a structured wiki rather than raw documents

Backlinks

Agentic Workflow Patterns

Augmented LLM is the building block that all five patterns compose
The orchestrator-workers pattern parallels the Brain-Hands Decoupling in Managed Agents — one brain, many hands
The evaluator-optimizer loop is structurally the same as the Agent Learning Loop — generate, evaluate, refine
CORAL implements a sophisticated version of orchestrator-workers with shared memory and asynchronous co-evolution
Hermes Agent's skill creation after complex tasks resembles the evaluator-optimizer: generate a solution, evaluate it, refine into a reusable skill

→

Building Effective Agents The foundational building block: an LLM enhanced with retrieval, tools, and memory. Current models actively use these capabilities — generating search queries, selecting tools, determining what to retain. Implementation should focus on tailoring capabilities to the use case and providing a well-documented interface. See Augmented LLM. →

Evaluator-Optimizer Pattern

Part of the Agentic Workflow Patterns taxonomy
The Augmented LLM is the building block — both generator and evaluator are augmented LLMs
Connects to Discrepancy Production vs Reduction — the evaluator produces discrepancy (identifying gaps), the generator reduces it (fixing them)
Goal-Setting Theory applies: the evaluation criteria function as specific, challenging goals that drive performance

→

AI Agents

Building Effective Agents — Anthropic's guide to composable agentic patterns: start simple, add complexity only when it helps
Agentic Workflow Patterns — taxonomy of five patterns: prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer
Augmented LLM — the foundational building block: LLM + retrieval + tools + memory
Agent-Computer Interface — ACI: designing tool interfaces for agents with the same rigor as HCI for humans
Orchestrator-Workers Pattern — central LLM dynamically decomposes tasks and delegates to worker LLMs
Evaluator-Optimizer Pattern — generator + evaluator LLMs iterating in a refinement loop

→