https://site.fredlab.dev/categories/architecture/Recent content in Architecture on FredHugoenCopyright (c) 2020-2024 ThuliteFri, 24 Jul 2026 08:00:00 +0200https://site.fredlab.dev/blog/excel-but-agent-ready-how-fred-turns-spreadsheets-into-queryable-knowledge/Fri, 24 Jul 2026 08:00:00 +0200https://site.fredlab.dev/blog/excel-but-agent-ready-how-fred-turns-spreadsheets-into-queryable-knowledge/Excel is not a document — it’s a data source wearing a document’s clothes. Here is how Fred rebuilds the structure of a workbook deterministically, hands the agent a global map plus queryable tables, and adds a server-side value locator that collapses a dozen blind queries into a single targeted one.https://site.fredlab.dev/blog/ingestion-shouldnt-be-a-choice/Wed, 10 Jun 2026 12:00:00 +0200https://site.fredlab.dev/blog/ingestion-shouldnt-be-a-choice/A look at why three ingestion modes create more problems than they solve, how the PDF extraction pipeline was rebuilt to be ten times faster with 90% fewer LLM calls, and the unified ingestion architecture that points toward removing the choice entirely.https://site.fredlab.dev/blog/tabular-data-in-fred-rebac-scoped-parquet-s3-and-duckdb/Thu, 16 Apr 2026 12:00:00 +0200https://site.fredlab.dev/blog/tabular-data-in-fred-rebac-scoped-parquet-s3-and-duckdb/Fred now treats tabular data like the rest of the corpus: document-scoped, team-aware, and authorization-first. This post explains why the old SQL-store model was limiting, how the new Parquet plus DuckDB runtime works, and what this changes for security, performance, and operations.https://site.fredlab.dev/blog/rich-mode-extending-rag-with-direct-visual-evidence/Mon, 13 Apr 2026 12:07:05 +0100https://site.fredlab.dev/blog/rich-mode-extending-rag-with-direct-visual-evidence/An exploration of the structural limitations of text-based RAG systems, supported by benchmarks, and how Rich mode introduces a multimodal retrieval strategy that preserves and leverages images as first-class evidence.https://site.fredlab.dev/blog/durable-ai-agents-orchestrating-the-future-with-fred-and-temporal/Wed, 04 Feb 2026 10:00:00 +0200https://site.fredlab.dev/blog/durable-ai-agents-orchestrating-the-future-with-fred-and-temporal/Fred extends its existing agentic backend with a Temporal-powered execution path for long-running tasks. Workflows remain resilient through retries, heartbeats, and durable state, while agent contracts and tool boundaries stay explicit and portable.https://site.fredlab.dev/blog/comparing-graph-centric-and-reflective-rag-agents-in-fred/Tue, 16 Dec 2025 12:07:05 +0100https://site.fredlab.dev/blog/comparing-graph-centric-and-reflective-rag-agents-in-fred/A comparison of Fred’s graph-centric RAG agent (Richard) and reflective RAG agent (Rico), including architecture and evaluation metrics.https://site.fredlab.dev/blog/integrating-external-a2a-agents-into-freds-backend/Fri, 12 Dec 2025 10:00:00 +0200https://site.fredlab.dev/blog/integrating-external-a2a-agents-into-freds-backend/Fred now supports A2A agents alongside MCP agents. Developers can deploy standalone, language-agnostic agents and register them in Fred as first-class citizens. A new agentic bridge handles protocol translation, streaming adaptation, UI rendering, and security integration—unlocking industrial-grade agent development.https://site.fredlab.dev/blog/mcp-vs-rest-for-rag-in-fred-why-we-run-both/Sun, 17 Aug 2025 10:00:00 +0200https://site.fredlab.dev/blog/mcp-vs-rest-for-rag-in-fred-why-we-run-both/Fred uses REST for graph-driven RAG (Rico, Rico Pro) and MCP for tool-driven RAG (Dominic). Running both lets us measure latency, reliability, governance, and developer ergonomics—without changing our UI or schemas.https://site.fredlab.dev/blog/ingest-smarter-not-harder-fred-gets-a-temporal-brain/Sun, 20 Jul 2025 19:00:00 +0200https://site.fredlab.dev/blog/ingest-smarter-not-harder-fred-gets-a-temporal-brain/Thanks to Temporal, Fred makes document ingestion as reliable as microservices — with automatic retries, fine-grained observability, and custom pipelines for push or pull ingestion. A true game-changer for AI apps.https://site.fredlab.dev/blog/expanding-fred-with-mcp-a-playground-for-next-gen-agentic-architectures/Mon, 19 May 2025 12:00:00 +0200https://site.fredlab.dev/blog/expanding-fred-with-mcp-a-playground-for-next-gen-agentic-architectures/Fred is now capable of serving and consuming components over the Model Context Protocol (MCP). Discover how this unlocks new architectures and makes Fred a live testbed for building interoperable AI agents.