标签：#rag

Before adopting DSPy, prove the LM program has a contract DSPy is easy to undersell. If you describe it as "a nicer way to write prompts", you will probably test the wrong thing. The better first test is this: can your language-model workflow be expressed as a small program with declared inputs, declared outputs, measurable examples, and an optimizer that is allowed to change prompts without changing the business boundary? That is the point where DSPy starts to make sense. The upstream README positions DSPy as a framework for programming, rather than prompting, foundation models. The current Doramagic project pack for stanfordnlp/dspy also points at the same starting command: pip install dspy . The package metadata I checked today declares name="dspy" , version 3.3.0b1 , and Python >=3.10,<3.15 . The GitHub API snapshot showed the repository was still active, with a push on 2026-07-05 and recent pull requests around empty evaluation sets, document formatting, and GEPA trace attribution. That activity is useful, but it is not the adoption proof. The proof is whether your own task can survive three separations. 1. Separate the task contract from the prompt In DSPy terms, the first artifact worth reading is not a clever prompt. It is the Signature. A Signature says what goes in and what must come out. A Module wraps that contract into something callable. An Adapter turns the contract into the provider-facing prompt format and parses the model response back into fields. That gives you a practical review question: Can the team name the output fields before anyone starts tuning wording? If the answer is no, DSPy will not rescue the workflow. You are still negotiating the task itself. The first pass should be a tiny contract: input text, retrieved context if needed, answer, citations, confidence, or whatever fields the product actually needs. Do not start with agents, tools, or multi-stage pipelines. 2. Separate compilation from runtime correctness DSPy optimizers can impr

If your memory is hosted, your thoughts are leased. We did not just move our files to the cloud. We moved our working memory. Andy Clark and David Chalmers called it the extended mind in 1998. The thesis was simple. Cognition leaks into the tools we trust. A notebook can be part of your mind if you access it reliably. In 2026, that notebook is a vector database owned by a platform with a legal department. Your extended mind now has terms of service, retention policies, and a compliance team that answers subpoenas faster than you answer email. I am not interested in nostalgia for paper. I am interested in architecture that preserves agency. The fix is not to think less with machines. It is to think locally with machines you control. That is why I built a second brain that lives on a Raspberry Pi 5 with NVMe and a Hailo-8 accelerator, running Retrieval Augmented Generation completely offline. No API keys. No telemetry. No third party that can be compelled to hand over your associative graph. This is the expanded blueprint. More cohesive, more rigorous, and more useful than the usual cloud versus local sermon. The extended mind, now with a landlord The original extended mind argument was about trust and coupling. If you reach for a tool as automatically as you reach for a memory, it counts as cognition. The cloud broke that coupling by inserting a landlord. Your retrieval is fast, but it is also observed, logged, ranked, and retained. Three consequences follow. First, epistemic pollution. When your queries train their models, your future answers are shaped by everyone else’s queries. Your private context gets diluted by the median user. Second, legal exposure. Your prompts, your uploads, your retrieval history, and your embeddings are business records. In many jurisdictions they are discoverable. You cannot plead the fifth for data you gave to a provider. Third, strategic fragility. A policy change, a price hike, a region block, and your cognitive prosthesis goes dark.

Have you ever noticed this? You explain your project to an AI chatbot, have a great conversation, then come back later... and it asks you to explain everything again. Start a new chat, and it's like meeting you for the first time. This isn't a bug. It's how most AI models work—they don't remember past conversations on their own. That's where Cognee comes in. Instead of making AI start from scratch every time, it gives AI a way to remember what matters. Why does AI forget everything? Most AI models don't have long-term memory. Every time you start a new chat, the AI only knows what you send in that conversation. It doesn't remember your previous chats, your project, or your preferences unless you provide them again. A common solution is RAG (Retrieval-Augmented Generation) . It stores your documents in a searchable database so the AI can look up relevant information when needed. RAG genuinely helps, but it only knows what sounds similar. It doesn't know "Priya" and "the payments lead" are the same person, or that this week's ticket shares a root cause with one from March. Similarity search finds neighbors — it doesn't understand relationships. That's where Cognee takes a different approach. What is Cognee? Cognee is an open-source memory layer for AI applications. Instead of making an AI start from scratch every time, Cognee helps it remember information across conversations. It can learn from your documents, files, websites, or notes and use that knowledge whenever it's needed. Unlike traditional RAG systems that mainly find similar text, Cognee also understands how different pieces of information are connected. That gives AI more accurate and meaningful answers. It's Apache-2.0, runs locally by default, and has 27k+ GitHub stars. How does it work? At a high level, the process is simple: flowchart LR A[Text, Files, URLs] --> B[remember()] B --> C[Cognee builds AI memory] C --> D[recall()] D --> E[AI answers using remembered knowledge] For example: "Alice bought a Pr

EDRSR — the Unified State Register of Court Decisions — is effectively all of Ukraine's judicial practice in open access. Today Qdrant holds **44M+ vectors : criminal (19M), civil (14.3M), commercial (5.1M), misdemeanors (5.6M). Vectorization of civil cases (CPC, justice_kind=1) — the largest cohort at 33.7M documents — runs on a dedicated EC2 instance (r6a.xlarge, 32 GB RAM, 2 TB gp3). Here's what's under the hood: models, pipeline, cost, rakes, and current status. Why Vectorize Courts When a lawyer searches "is there case law on recovering bank prepayment fees" — they don't want to open 40 decisions and read them through. They want the system to surface the top 5 most relevant ones, pull out key paragraphs, and show how courts reasoned. Full-text search (FTS) over keywords doesn't give that — it returns every document containing the word "fee", and there are thousands. For this semantic task you need vector representations of text. The model turns a paragraph from a decision into a point in a 1024-dimensional space; semantically similar paragraphs sit near each other. A kNN search in Qdrant returns the top K nearest, and an LLM composes the answer from exactly those relevant fragments. The only problem: the register is big. Very big. Scale Our prod database holds full texts of decisions starting from 2006. Breakdown by procedural type: Civil (CPC) — 33.7M documents. The largest category. Consumer, housing, labor, family. Criminal (CrPC) — 12M+ Administrative (CAS) — 14M+ Commercial (CC) — 6M+ Misdemeanors (CUaP) — 6M+ The Qdrant collection edrsr_decisions on a dedicated EC2 currently holds 44M+ vectors (122 segments, on_disk=true): | Proceeding type | justice_kind | Vectors | |—|—|—| | Criminal (CrPC) | 2 | 19,036,347 | | Civil (CPC) | 1 | 14,328,427 | | Misdemeanors (CUaP) | 5 | 5,579,432 | | Commercial (CC) | 3 | 5,098,662 | | Total | | 44,042,868 | Civil cases processed: 14.3M out of 33.7M — that's 42%. After CPC completes there will be roughly 63M+ vectors in

Local LLM Deployment, Agent Handbook, & LLM Cost Reduction: Applied AI Workflows Today's Highlights This week's highlights cover practical guides for running state-of-the-art LLMs locally and building AI agents, alongside an innovative technique to significantly cut LLM API costs for code processing. These resources focus on actionable insights and frameworks for real-world AI application development. Jamesob's guide to running SOTA LLMs locally (Hacker News) Source: https://github.com/jamesob/local-llm This GitHub repository provides a comprehensive, hands-on guide for setting up and running state-of-the-art Large Language Models (LLMs) on local hardware. It meticulously covers the necessary tooling, dependencies, and configuration steps required to get various open-source LLMs operational without relying on cloud APIs. The guide emphasizes practical considerations for local inference, including hardware requirements, model quantization techniques, and performance optimization for different architectures, directly addressing production deployment patterns. It serves as an invaluable resource for developers and researchers looking to experiment with LLMs, develop applications offline, or reduce costs associated with cloud-based inference by leveraging local compute. The guide offers concrete details and actionable steps, making it an essential resource for anyone aiming to implement LLMs in a controlled, private, or cost-effective environment. Comment: This guide is fantastic for anyone wanting to get serious about local LLM development. It covers the nitty-gritty details of setting up your environment and getting models like Llama-3 running efficiently on consumer hardware, which is crucial for privacy and cost savings. 60% Fable cost cut by converting code to images and having the model OCR it (Hacker News) Source: https://github.com/teamchong/pxpipe The pxpipe project introduces an innovative technique to drastically reduce API costs when processing code with Lar

The $149 Dune keyboard can be a meeting controller at least and a script-executing keypad at best.

Applied AI: Copilot's Kimi K2.7, AI Agent Workflow Barriers, Open-Source Life Planner Today's Highlights This week's top AI news covers a significant upgrade to GitHub Copilot with the Kimi K2.7 Code model, enhancing developer productivity through advanced code generation. We also explore the practical challenges faced by AI agents in fully automating workflows due to "last mile" integration issues, alongside a hands-on look at a new open-source AI life planner that demonstrates real-world application of AI tools. Kimi K2.7 Code is generally available in GitHub Copilot (Hacker News) Source: https://github.blog/changelog/2026-07-01-kimi-k2-7-is-now-available-in-github-copilot/ GitHub Copilot has integrated the Kimi K2.7 Code model, making this advanced code generation capability generally available to its users. This update signifies a continuous improvement in the underlying AI models that power development tools, specifically in the domain of code generation and assistance. Kimi K2.7, presumably an internal or specialized model from GitHub's AI research, focuses on enhancing the quality, relevance, and efficiency of generated code suggestions, auto-completions, and code explanations within the Copilot environment. For developers, this means a more accurate and helpful programming assistant that can better understand context and intent. The deployment of Kimi K2.7 into a widely used production tool like GitHub Copilot demonstrates a key pattern in applied AI: iterating on foundation models and integrating improved versions directly into developer workflows. This enhancement aims to boost developer productivity by reducing the time spent on boilerplate code, debugging, and searching for solutions, allowing engineers to focus on higher-level architectural and design challenges. This release confirms the ongoing progress in AI's capability to augment the software development lifecycle. Comment: New model, better code generation – straightforward for Copilot users. This

You built a system to surface information so a person could decide. Somewhere it started deciding for them — the output stopped saying "here's what the documents show" and started saying "you should do X." Nobody designed that drift. An LLM, when asked a question, produces an answer-shaped thing, and an answer easily becomes a verdict. What everyone tries A prompt instruction: "Don't make recommendations." "Only state what's in the documents." People add the line and assume the boundary is enforced. Why it doesn't work A prompt instruction is a request, not a guardrail. The model follows it most of the time, then on the input that matters produces a confident recommendation anyway, because nothing structurally prevents it. "Please don't make recommendations" is to a guardrail what a sticky note saying "please don't enter" is to a locked door. And the stakes are higher than they look. When output drifts from evidence to verdict, accountability moves. As long as the system returns evidence and a human decides, the human owns the decision. The moment the system returns a verdict and the human defers, the system is deciding things it was never validated to decide — and when one is wrong, accountability is a blank. High-stakes fields separate evidence extraction from judgment on purpose; most RAG systems erase that line by default. The one shift Decide what the output is and enforce it structurally. An output should declare itself: answer, evidence, missing facts, or out-of-scope. "Return decision material, not a decision" has to live in the output contract and in gates — not in a polite request to the model. The system supplies frames; the human supplies verdicts. This is the output boundary — one of three places production RAG dies. Read the full version on my blog , where this connects to the RAG Failure Diagnosis Kit for teams debugging production RAG.

Honda wants in on the lucrative energy storage market. This week it began producing batteries destined for data centers, not driveways.

Venice AI is already profitable, with annualized run-rate revenues of over $70 million, CEO Erik Voorhees said.

If you run a RAG system in production, you eventually hit a dilemma that has nothing to do with your model and everything to do with your cache. Cache the answers to save tokens and latency, and one day a source document changes — but your cache keeps cheerfully serving the answer it built from the old document. Nobody gets an error. The number is just quietly wrong. Cache nothing , and every single call re-retrieves the same chunks, re-reads them, and re-pays the full context bill to rebuild an understanding you already built five minutes ago for a nearly identical question. Stale or expensive. Most teams pick "expensive" because at least it's correct, then bolt on a TTL and hope. This post is about why the TTL doesn't save you, and about two specific, mechanical fixes that let you cache RAG context and stay fresh. I maintain an open-source library called Coalent that implements both, so I'll use it for the runnable examples — but the two ideas are portable and worth stealing even if you never pip install anything. Failure mode 1: the stale RAG cache (and why a TTL won't save you) Here's the standard "answer cache" sitting in front of retrieval: answer = cache . get ( query ) if answer is None : chunks = retriever . retrieve ( query ) answer = llm . synthesize ( query , chunks ) cache . set ( query , answer , ttl = 3600 ) return answer This works until billing.md changes. The refund window goes from 30 days to 14. Your cache has an answer keyed on "what is our refund policy?" that says 30, and it will keep saying 30 for up to an hour — or forever, if the same question keeps refreshing a TTL that never expires under load. The reason this is hard is that the cache key (the query) has no relationship to the thing that changed (the source). You cached an answer; you threw away the fact that this particular answer was derived from billing.md . So when billing.md changes, you have no way to find the answers that depended on it. The TTL is a confession that you can't answ

I have been building AI products for a while and kept running into the same problem. Every project that involves querying documents with AI requires the same foundation before you can build anything interesting: a chunking strategy, an embedding pipeline, a vector database, re-ingestion logic when content changes, and a retrieval layer on top. It is not hard, it is just a lot, and it is not the part you actually want to be building. So I built Kognita to handle it as a managed API. You push content in via API, text or files, and get back hybrid search over a knowledge base. Kognita handles chunking, embedding, indexing, and automatically re-embeds when you update content. It is opinionated: we pick the embedding model and chunking strategy. The trade-off is less flexibility for a much faster path to a working knowledge layer. What we are looking for We want 10 teams who are building something that needs a knowledge layer and are willing to test it honestly. The ask is: what broke, what was confusing, what you needed that was missing. Not looking for compliments. Looking for people who will actually use it and tell us where it falls short. What you get Unlimited knowledge bases 10 GB storage 100 GB egress per month 50 GB file storage No credit card. No time limit. Higher than our standard paid plan. Who it is for Engineering teams building AI features over documents who do not want to manage the underlying infrastructure themselves. If you need full control over your embedding models or retrieval strategies, this is probably not the right fit. If you want to skip the pipeline and get to building, it might be. How to get started Sign up at kognita.io. Drop a comment here if you sign up and I will make sure you are on the early adopter tier.

Originally published on PrepStack . Everyone's first RAG pipeline is the same four boxes: documents, chunk, vector DB, LLM. It demos in an afternoon and then quietly betrays you in production — stale answers, no relationships, no governance, and a model guessing from fragments. The fix is not a bigger vector index. It is to stop storing documents and start storing knowledge . That is Open Knowledge Format (OKF). To be clear up front, because the title is deliberately provocative: OKF does not kill embeddings. Vectors still do the recall. What OKF kills is blind chunking — slicing opaque documents into context-free fragments and hoping cosine similarity reassembles meaning. On Mattrx , a multi-tenant marketing-analytics SaaS (.NET 9 + Azure SQL + a Python FastAPI AI service), replacing blind chunking with OKF + a Context Engine took the assistant's hallucination rate from 18% to 3% and stale-answer rate from 11% to 1.5% . TL;DR Dimension Documents → chunk → vector DB (before) OKF + Context Engine (after) Unit of knowledge Opaque chunk of text Typed, governed knowledge unit Structure None — chunks are islands Metadata + relationships + schemas Freshness Snapshot, rots silently valid_until + live API refs Rules Buried in prose, ignorable First-class data the engine enforces Retrieval Top-k cosine Hybrid + vector + graph Multi-hop questions Unanswerable Answered via relationships Results after the rebuild: Knowledge base restructured into ~11,000 OKF units (Markdown + metadata + relationships + APIs + schemas + business rules). Hallucination 18% -> 3% ; faithfulness 0.96 ; answer-relevance 0.91 . Context tokens/call 14k -> 3.5k — structure lets the engine attach the right thing, not everything. Outdated-answer rate 11% -> 1.5% ( valid_until + metadata freshness). Multi-hop questions unanswerable -> answered via graph retrieval. Deprecated-plan recommendations recurring -> 0 (business rules enforced as data). The one mental shift: a chunk is a fragment of text with no id

Traditional content pages are often designed for a linear reader. The introduction sets context, the middle develops the idea, and the conclusion ties everything together. AI retrieval does not always work that way. A system may identify smaller content units, pull the most relevant section, compare it with other sources, and use that fragment to support an answer. The full page still matters, but the retrievable blocks inside the page matter just as much. A useful Tumblr post explains the idea in simple terms: https://www.tumblr.com/digitalisedsoul/820825642809573376/ai-does-not-read-your-content-like-a-human?source=share For Dev Community readers, the pattern is familiar. Poorly structured inputs lead to weaker outputs. If content is dense, vague, or dependent on surrounding paragraphs, it becomes harder to extract and reuse. If content is modular, clear, and properly scoped, retrieval becomes easier. Marketing teams can learn a lot from this. A strong content page should behave like a set of well labelled components. Each section should answer a specific question. Headings should be descriptive, not decorative. Paragraphs should avoid vague references such as the above point or this approach when the section may be read independently. Definitions should appear close to the terms they explain. Examples should include enough context to stand alone. Proof should be written as text, not only displayed as graphics. Internal links should connect related concepts in a way that helps both readers and systems understand the topic map. A page about AI search visibility, for example, should not only include one broad explanation. It should break the topic into useful blocks: what AI visibility means, why AI systems retrieve passages, how source trust works, what makes content reusable, and how brands should measure answer presence. Each block becomes a possible answer unit. That structure does not weaken the reader experience. It improves it. Developers, marketers, and busi

The AI Implementation Process I Use With Every Client Most AI projects do not fail at the model. They fail in the six weeks before anyone writes a prompt, and in the six weeks after the demo lands in a Slack channel and nobody knows who owns it. I have run enough of these now (from one-off automations to multi-agent content systems running unattended) that the process has converged into something stable. This is the version I actually use. It has five phases: scoping, POC, integration, evaluation, operations. Each phase has an exit criterion. If we cannot meet the exit criterion, we do not move forward. That single rule has saved more projects than any clever architecture choice. Phase 1: Scoping (1 to 2 weeks, fixed price) Scoping ends with a written document that names the workflow being automated, the system of record it touches, the success metric in hours or dollars, the data we have access to, and the smallest possible first slice. No model is chosen yet. No code is written. If we cannot produce that document, the engagement stops here and the client keeps the document. The hardest part of scoping is resisting the urge to solve the interesting problem. Clients almost always describe the AI-shaped fantasy ("an agent that handles all support tickets") when the real opportunity is narrower and uglier ("triage tier-1 tickets that mention billing, route to the right queue, draft a reply for human approval"). The narrower version ships. The fantasy does not. I run scoping as three sessions: Workflow walkthrough. Someone who actually does the work shows me their screen for an hour. I record it. I take timestamps. The point is to find the moments where a human is doing pattern matching that an LLM can do, and to find the moments where they are doing judgment that an LLM should not do. Data audit. Where does the input live? Where does the output need to go? What is the auth story? If the data is locked inside a SaaS product with no API and no export, that is the projec

You ask the AI for a bibliography. It hands you a title, authors, a journal, a year, a well-formed DOI. Everything is plausible, everything is clean. And one reference in two doesn't exist. Not "approximate": nonexistent. The DOI resolves to nothing, the paper was never written. The reflex is to ask the model again: "are you sure this source is real?" It says yes. Always. You just asked the forger about the authenticity of his forgery. Hallucination is plausible by construction An LLM doesn't store a database of publications. It generates likely sequences of words. A citation, to it, is a shape: a surname, an initial, two more names, a capitalized journal, a recent year, ten DOI digits. It produces that shape perfectly, because that's exactly what it's good at. The content doesn't need to be true to be plausible, it just needs to resemble. That's why a hallucinated reference is so vicious: it doesn't look like an error. A wrong calculation jumps out. An invented citation looks like a real one, until you click. Don't ask the culprit The golden rule fits in one sentence: never ask the model that hallucinated a citation whether that citation is real. For two reasons that compound. First, it doesn't have the information: it has no access to a registry, it can only regenerate something plausible. Second, even if it doubted, its self-evaluation bias pushes it to confirm what it already produced. You get a "yes" worth nothing. Verification has to come from elsewhere. From a source the model neither controls nor can invent: a metadata API. Three filters: existence, credibility, fidelity In my pipeline for writing technical dossiers, no reference enters the document before clearing three filters, in this order. Existence. The DOI must resolve. It's binary, and it's free. Crossref exposes its whole database: curl -s "https://api.crossref.org/works/10.1145/3290605.3300233" \ | jq '.message.title[0], .message.author[0].family, .message["published"]' If the API returns a title a

What This Guide Covers When you start building LLM-powered applications, one pattern becomes unavoidable: RAG (Retrieval-Augmented Generation) . LLMs only know what they were trained on. Your company's internal documents, the latest spec sheets, project-specific information — none of that exists in the model. To handle data the model doesn't know, you need a system that retrieves relevant knowledge in real time and injects it into the context. That's RAG. In this guide, we'll implement a RAG system from scratch using pgvector and Gemini, then extend it step by step through Tool Use, AI Agents, MCP, and cloud deployment. Step 1: Embedding · Vector DB · RAG — core implementation Step 2: AI Architect perspective — design decisions explained Step 3: Tool Use — LLM autonomously searches the DB Step 4: AI Agents — combining multiple tools Step 5: MCP — exposing tools as a server Step 6: Cloud deployment — Render × Supabase Three Concepts to Understand First Embedding Computers can't measure "semantic similarity" from raw text. Embedding converts text into a list of numbers (a vector), and semantically similar words produce numerically similar patterns. "dog" → [ 0.82 , 0.75 , 0.10 , ... ] 768 numbers "cat" → [ 0.78 , 0.72 , 0.12 , ... ] ← similar pattern to "dog" "bank" → [ 0.08 , 0.10 , 0.85 , ... ] ← completely different Gemini's embedding model handles this conversion. Vector DB A regular DB searches by keyword matching. A vector DB searches by numeric distance — meaning it finds semantically related documents even when the exact words don't match. -- Regular search (misses if keywords don't match) SELECT * FROM docs WHERE body LIKE '%F1 score%' ; -- Vector search (finds semantically related docs) SELECT * FROM docs ORDER BY embedding <=> query_vector LIMIT 3 ; Search for "how to measure model performance" and it finds "F1 score calculation" — even without matching words. We use pgvector , a PostgreSQL extension, for this. RAG LLMs are limited to their training data. R

When confronted with cancer, Connor Christou fed everything tied tied to his regime — blood results, scan data, wearable output, journal entries — into Claude.

Mitigating Hallucinations in Theology AI: Implementing Groundedness Evaluation Pipelines For software developers and indie hackers, the era of building generic wrapper APIs is over. The real value now lies in highly specialized, niche vertical applications. One of the most fascinating, complex, and underserved niches is the intersection of artificial intelligence and religious doctrine. Building a catholic ai tool presents unique software engineering challenges. Unlike general-purpose chatbots, a theology ai application cannot afford to "hallucinate" or generate creative interpretations of established doctrines. In this space, an inaccurate answer is not just a software bug; it is a theological error. To build a high-quality, trustworthy catholic ai app , developers must move past basic prompt engineering. We must implement robust groundedness evaluation pipelines. This article explores the technical journey of building a specialized catholic ai chatbot , the catholic church stance on ai , our choice of tech stack, and how to build a production-grade groundedness pipeline to keep your AI aligned with official church teachings. The Catholic Church Stance on AI: Designing for Ethics and Trust Before writing a single line of Dart, Swift, or Python, we must understand the ethical landscape of ai and theology . The Vatican has taken an surprisingly proactive approach to artificial intelligence. Pope Francis has frequently spoken on the topic, advocating for "algor-ethics"—the ethical development of algorithms. The catholic church stance on ai emphasizes that technology must serve human dignity and remain aligned with truth. ┌─────────────────────────────────┐ │ The Vatican's Algor-ethics │ └────────────────┬────────────────┘ │ ┌─────────────────────────┴─────────────────────────┐ ▼ ▼ ┌──────────────────┐ ┌──────────────────┐ │ Human Agency │ │ Doctrinal Truth │ │ AI must assist, │ │ AI must not alter│ │ never replace │ │ established dogma│ └──────────────────┘ └─────────

Most discussions around MCP focus on tool calling. That is natural. When people first see MCP, the obvious use case is simple: Let the AI call external tools. A model can read a GitHub issue. A model can query a database. A model can update a file. A model can call an API. In that sense, MCP looks like an RPC layer for AI agents. That is useful. But I think it may not be the most important use of MCP. The more interesting use is this: MCP can distribute context, rules, skills, and operating contracts to AI clients. In other words, MCP is not only a way for AI to call tools during work. It can also be a way to define the working environment before the work starts. The problem with RAG RAG is usually used to answer this question: What information might be relevant to this request? The system searches documents, retrieves chunks, and gives them to the model. This works well for many cases. But it has structural limits. RAG retrieves likely relevant information. It does not necessarily define how the work should be done. For team-level AI work, this is a problem. A team does not only need information. A team also needs shared rules. For example: What is the authoritative source? What should be treated as unknown? When should the AI stop? When is human confirmation required? What is the closure condition? Which workflow should be used? Which domain skill applies? What evidence must be recorded? RAG can retrieve documents that describe these rules. But retrieval is not the same as governance. A retrieved chunk is just context. It is not necessarily an operating contract. The problem with local prompts Many teams try to solve this with prompts. They write instructions like: Follow our coding rules. Use this design document. Ask questions when unclear. Do not make risky changes. This helps, but it does not scale well. Each developer may have a different local prompt. Each AI client may load a different file. Each repository may contain a slightly different version of the ru