By Ryan Watts — 27 Feb 2026

Agentic tools and frameworks (by purpose)

Marvin AI (Pydantic AI) — Custom Python agents (e.g. data analyst, data validation, data aggregation) via marvin.run and agents parameter; marvin.summarize for summaries; used for specialized and anonymous agent solutions.
Marvin Agentkit (concept/implementation) — Load agents and skills from Markdown, compose prompts, build Marvin Agent instances, and bind tools from a registry; aligns Cursor-style agent/skill conventions with Marvin runtime.
Pydantic — Structured inputs/outputs and validation for agents (e.g. table definitions, SQL examples, validation rules, agent configs).

Prefect — Flow and task orchestration, schedules, retries, artifacts, and events; runs daily workflows that call external agents and LLMs; used to orchestrate multi-step agent pipelines.
Langflow — Visual workflow builder and custom components (e.g. auth gate, consent checker, PHI redactor, input/output encryptors, audit logger) for compliance-oriented, workflow-driven agents; deployment to a hosted Langflow instance.
n8n — Workflow automation (Helm-deployed); webhooks and execution config for higher-level, workflow-driven automation alongside other agent stacks.

MindsDB — Hosted agents, federated queries, natural-language-to-SQL and forecast/smart-action generation; prompt templates and model config; synced with external DB and OpenAI-generated content.
OpenAI — Tool-calling loops in copilot-style agents; GPT models for smart actions, summarization, and embeddings.
Anthropic / OpenAI (in Obsidian) — Model provider for title generation, post-processing, and model selection in a vault/AI setup.

Hierarchical rules engine — structured and unstructured information in a DB; get_active_rules / get_rules_for_prompt filter by scope and format rules for LLM prompts so operational constraints are enforced in agent outputs.
Trigger-based analysis rules — Documented rules for when to run analyses (e.g. weekly WoW, monthly MoM, channel breakout, labor/discount); drives which analyses agents run and when.
Structured validation rules — Validation categories (e.g. data quality, format, integrity, PII) and rule models (condition, logic, status); used by a validation agent to check data and report pass/fail/na.

MCP (Model Context Protocol) — Cursor (e.g. MCP filesystem), Notion Busy Brain (search, fetch, create/update pages and DBs), Langflow streamable HTTP (tools/list, tools/call), and semantic-vault MCP in Obsidian; extends agents with shared tools and knowledge.
Langflow MCP integration — Streamable HTTP client so a copilot can discover and call Langflow workflow tools; configurable endpoint and API key; merges Langflow tools with local tools in one agent.

Notion (Busy Brain MCP) — Semantic search over workspace and connected sources (Slack, Google Drive, Jira, Linear, etc.); fetch/create/update pages, databases, comments; used as knowledge and collaboration layer for agents.
Supabase — Rules storage, smart-action records, and consent/audit data for compliance and agent context.

CrewAI — Multi-agent “crews” and flows in a sandbox; experimentation with crew-based agent patterns.

Cursor / Cline — Rules, skills, worktrees, memory bank; plugin quality gates; create-rule and create-skill flows; editor-integrated agent authoring and execution.
SystemSculpt (Obsidian) — AI-in-vault with OpenAI/Anthropic; MCP (e.g. filesystem); Marvin framework referenced in extractions and settings.

Cryptography (Fernet) — AES-256 in Langflow custom components (input/output encryptors) for PHI and compliance (e.g. HIPAA-oriented) before/after processing.
Custom Langflow compliance components — Auth gate, consent checker, PHI redactor, encryptors, audit logger, error handler, output validator; 12-component compliance framework for regulated AI workflows.

FastMCP — MCP server config, Cursor/Claude install, and tooling used alongside Marvin-based agent projects.