Agentic Frameworks MOC

This map covers the mechanics of how autonomous agents reason, act, and interact with knowledge bases.

Parenthetical tags such as Freedom, Control, Handoffs, and Retrieval are local organizing mnemonics for navigation inside this vault. They are not protocol-standard or vendor-endorsed classifications.

The Knowledge Maps

hf-agents-course-moc: Theoretical backbone from the Hugging Face course.
mcp-moc: Comprehensive guide to the Model Context Protocol.

The Thought-Action Loop

agent-thought-cycle: The core Thought-Action-Observation cycle.
react-pattern: Tactical interleaving of reasoning and acting.

Mechanics of Agency

agent-tools: How to define and provide capabilities to an LLM.
agent-actions: The "Stop and Parse" loop and tool usage.
code-agents: The shift from JSON schemas to executable logic.

Anthropic & Claude Ecosystem

anthropic-moc - Comprehensive guide to the Anthropic platform and agentic features.
anthropic-messages-api: Direct Claude API request/response semantics and token counting.
anthropic-xml-prompt-structuring: XML-style prompt boundary markers for separating instructions, context, examples, and live input inside Claude text prompts.
anthropic-tool-use: Anthropic-specific client/server tool loop behavior.
anthropic-agentic-loop: Higher-level loop model: client tools, server tools, and pause_turn.
anthropic-server-tools: Server-executed web, fetch, code-execution, and tool-search behavior.
anthropic-streaming-patterns: SSE event flow, streamed tool arguments, and thinking block streaming.
anthropic-adaptive-thinking: Model-driven reasoning allocation via effort; interleaved thinking for agentic workflows.
anthropic-tool-runner-sdk: SDK tool loop automation with compaction for long-running agents.
anthropic-mcp-connector: Messages API as MCP client — remote servers without local MCP client.
anthropic-managed-agents-model: Hosted agent runtime; Agent/Environment/Session state machine.

Communication & Formatting

chat-templates: Managing System, User, and Assistant roles.
anthropic-xml-prompt-structuring: Provider-specific prompt markup pattern for making mixed prompt regions legible to Claude.
anthropic-xml-tags-cheat-sheet: Compact prompt-authoring reference for Claude XML-tag layouts, few-shot wrappers, and multi-document containers.
yanp-for-agentic-workflows: Why strict metadata is required for agentic parsing.

Frameworks & toolkits

hermes-moc: Dedicated map for the Hermes cluster — foundations, literature, and derived runtime specs.
hermes-agent: Full agent environment spanning tools, skills, memory, gateway surfaces, and durable background execution.
hermes-vs-adk-openai-agents-langgraph: Comparative positioning note for when the question is not "what can agents do?" but "what category of system is Hermes relative to ADK, OpenAI Agents SDK, and LangGraph?"
hermes-provider-abstraction: Provider-agnostic substrate layer — declarative provider profiles, runtime switching, and auxiliary routing.
hermes-prompt-assembly: Prompt-control plane that compiles identity, environment, skill, and project context into the live system prompt.
hermes-profiles: Identity partitioning layer for running multiple independent Hermes agents on one install.
hermes-skills-system: On-demand procedural knowledge substrate built around SKILL.md and progressive disclosure.
smolagents: Hugging Face's code-first agent library (Freedom).
openai-swarm: Experimental multi-agent orchestration pattern centered on handoffs.
openai-symphony: Service specification for background coding-agent orchestration.
openai-symphony-orchestration-state-machine: Internal claim states, retries, reconciliation, and bounded worker attempts.
openai-symphony-workflow-contract: WORKFLOW.md as repository-owned control artifact for prompt, config, hooks, and policy.
openai-symphony-trust-boundary: Trusted-environment framing, implementation-defined safety posture, and harness hardening guidance.
openai-agents-sdk: Production-oriented OpenAI agent SDK for scalable orchestration.
hermes-vs-openai-symphony: Comparative note contrasting tracker-native orchestration with Hermes's persistent agent environment.
adk-moc - Google's code-first toolkit for complex agent systems (ADK).
llamaindex: Data-oriented agent toolkit with strong retrieval and workflow support (Retrieval).
langgraph: Stateful orchestration framework with explicit control-flow graphs (Control).
pydantic-fastapi-agents: Using Pydantic for robust tool schema definition.
alternative-agent-frameworks: Exploring CrewAI and Microsoft AutoGen.

Advanced operational patterns

multi-agent-systems: Orchestrating specialized teams of agents.
orchestration-tradeoffs: Comparing Swarm (Freedom) vs. ADK (Control).
agentic-rag: Autonomous retrieval and self-correction.
modular-rag-hub: Adaptive Routing, Self-RAG critique loops, and GraphRAG hybrid patterns — the full Modular RAG stack.
graph-orchestration: Modelling behavior as deterministic workflows.
chromadb: Open-source embedding database for semantic memory.
hermes-bounded-memory: Practical bounded-memory pattern — tiny always-on prompt memory plus transcript retrieval.
hermes-gateway: Messaging-daemon pattern that exposes the same agent runtime across many platforms.
hermes-command-control-plane: Operator grammar for steering the runtime itself through slash commands, gateway controls, and dynamic command registration.
hermes-context-compression: Lineage-preserving compaction protocol that rotates session IDs while keeping a logical conversation continuous.
hermes-tool-registry: Registry/toolset/MCP integration layer that governs which tools the model can actually see and call.
spec-hermes-agent-loop: Derived execution contract for Hermes turns — prompt assembly, tool loops, memory snapshot semantics, and surface overlays.
hermes-cron: Scheduled autonomous execution layer for fresh-session agent runs and script-only watchdogs.
hermes-subagent-delegation: Isolated child-agent fan-out inside a single parent turn.
hermes-kanban: Durable multi-agent board for named profiles, resumable handoffs, and cross-time collaboration.

community-report-generator: Automated synthesis of emergent communities via embedding-link hybrid weighting.
telephony-agents: Autonomous interactions with PSTN and mobile networks; AgentPhone and streaming voice patterns.

Training & Fine-Tuning

function-calling: Moving from prompted to learned native agency.
anthropic-prompt-caching: Provider-specific context reuse for repeated prompt prefixes.
lora: Lightweight adaptation for agentic specialized tasks.

Observability & Evaluation

agent-observability: Logs, metrics, and traces — the instrumentation layer that makes production agents legible. Start here for the full signal taxonomy and framework routing.
agent-evaluation: Offline benchmarking and online production monitoring loop.
llm-as-a-judge: Automated grading using separate LLMs.
gaia-benchmark: Prominent benchmark for evaluating general AI assistants.

Protocols & Infrastructure

agentic-protocols: The emerging standards (MCP, A2A) for interoperability.
spec-agentic-source-orchestrator: The "Knowledge Compiler" for automated vault ingestion.
spec-firecrawl-pgvector-pipeline: Infrastructure for high-fidelity web scraping.
firecrawl-api-v2-reference: Technical specs for the Firecrawl integration.

synthesis-intelligence-layer: Local vault conventions for epistemic risk tiers (T0-T5) and arbitration protocols for autonomous synthesis.

Platform & Runtime SDKs

dotnet-agent-integration: Bridge note — architectural patterns for .NET in the agent loop.
dotnet-moc: The .NET ecosystem for high-performance agentic backends.
csharp-moc: Building Tier-1 agents with the C# language.
foundry-local: Microsoft's hardware-optimized local inference SDK.
local-agent-environments: Running models locally via Ollama and LiteLLM.
docker-sandbox: Secure environment isolation for agents.
hardware-aware-inference: Runtime and format selection for local inference — CUDA, MLX, DirectML, ONNX, and NPU targets; routes to Ollama, Foundry Local, and LM-Kit.NET.

Practical Use Cases

gala-agent-use-case: Building a multi-tool gala assistant.
agentic-rag: Implementation patterns for dynamic data retrieval.

Specialized Domains

agents-in-games: Autonomous NPCs and emergent storytelling.
pokemon-battle-agent: case study in turn-based environment interaction.

Knowledge Vault Operations

vault-audit-tool-spec: Formal schemas for local maintenance utilities.
wiki-pattern-operations: Ingest, Query, and Lint cycles.
wiki-as-codebase: Treating knowledge as a manageable software project.

lit-llm-wiki: Foundational literature for the Compounding Artifact and Wiki-as-Codebase patterns.
lit-foundry-local: Documentation for the Azure AI Foundry Local SDK.
lit-language-summaries: Core patterns for Python and Racket development.

Core Paradigms & Theory

High-Level Architectures

cognitive-architectures - Reasoning models and memory-augmented agents.
daemon-design-pattern - Persistent, backgrounded agent processes.
hierarchical-graph-synthesis - Multi-layered graph building and arbitration.
community-polyglot-agent-platform - Vision for multi-language agent coordination.
lattice-implementation-guide - Practical steps for building capability lattices.