Recipes

Build intelligent document search using text embeddings for meaning-based retrieval

Transcribe audio content and analyze transcripts with AI functions

Extract metadata, scenes, and insights from video content using AI functions

Search across images, videos, audio, and documents using unified vector search

Build a recommendation system using vector similarity and user behavior

Create automated media processing pipelines with status tracking and notifications

Build comprehensive analytics for media usage, engagement, and performance

Track media compliance, retention policies, and audit trails for governance

Synchronize media assets across multiple platforms and track sync status

Build an AI agent that learns from thumbs-up/down feedback — train a model on past interactions to predict whether a draft reply will satisfy the user, all in one SQL database. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

An incident-triage agent that correlates the alert with recent deploys, runbooks, and customer impact — producing a structured 'here's what I see / suspected cause / recommended action' in one SQL call.

A compliance-reviewer agent that walks the transaction table, consults KYC docs and regulations by meaning, surfaces hits with citations, and drafts an SAR-style write-up — all in SQL via AGENT_RUN.

A clinical-assistant agent that grounds answers in the patient's chart AND the relevant guideline — never just one or the other. Built on a single SQL session with AGENT_RUN.

An HR-assistant agent that combines the employee's specific row (country, tenure, role band) with the right policy doc to answer correctly — not a generic RAG-only bot that hallucinates country-specific rules.

Stand up an in-database claims-adjuster agent that retrieves policy clauses by meaning, scans prior-claim history for fraud signals, and recommends an action — all in a single SQL session via AGENT_RUN.

A contract-reviewer agent that diffs an inbound MSA against your standard playbook AND your recent negotiated precedents — surfacing deviations with the closest prior deal as a reference.

A shipment-exception agent that joins the late shipment with the customer's SLA tolerance and recalls similar past exceptions, then drafts both the customer notice and the carrier dispatch — in one SQL call.

A returns-processing agent that reads the order, checks return-policy by meaning, scans for abuse patterns in prior returns, checks live inventory for swap availability, and recommends a specific action — all in SQL via AGENT_RUN.

A tier-1 support agent that reads the customer's actual account state AND the help docs in one call, then either resolves the ticket or escalates with full context — not a RAG-only chatbot.

An account-research agent that pulls the CRM record, finds similar past wins by meaning, and outputs a one-page briefing before the sales call — all in SQL via AGENT_RUN.

Build a complete AI customer-support agent on one database — semantic ticket memory, RAG over your help docs, intent routing, priority scoring, and a customer graph, no five-service stack. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Build a personal AI assistant on one database — long-term memory of preferences, semantic tool routing, RAG over your notes, and grounded replies. No five-service stack. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Build an AI research agent that reads documents, builds a knowledge graph, answers multi-hop questions with citations, and synthesizes a briefing — RAG plus a knowledge graph in one database. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Build an AI triage agent that classifies every incoming request by meaning, scores its urgency with AutoML, and routes it to the right queue or human — all on one database. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Turn unstructured text into a queryable knowledge graph an AI agent can reason over — extract entities and relationships with GENERATE, store them as a Cypher graph, all in one database. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Build an AI SDR agent on one database — score leads with AutoML, match them to case studies by meaning, map the buying committee as a graph, and draft personalized outreach with GENERATE. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Store a chatbot's full conversation history and keep a rolling summary so it never blows the context window — turn buffer, semantic recall, and auto-summary in one SQL database. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

The next-step variant of the rolling-summary recipe — replace the single prose summary with a typed chat_facts table (category, key, value, confidence, updated_at). Higher recall, surgical updates, no contradiction accumulation.

Give an AI agent two memory types — episodic (what happened, when) and semantic (durable facts) — in one SQL store, and recall across both by meaning. Framework-agnostic: Claude Code, OpenClaw, LangChain, or a voice agent.

Add persistent, semantic long-term memory to any AI agent — store and recall memories by meaning with vector embeddings in plain SQL. Framework-agnostic: works with Claude Code, OpenClaw, LangChain, or a custom or voice agent.

Give an AI agent multi-hop reasoning — combine semantic retrieval with a knowledge graph so it can answer questions that need several connected facts, not just one passage. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Classify what a user wants — billing, support, sales, cancel — so your AI agent routes the conversation correctly, using vector nearest-example matching in plain SQL. No training job, works with Claude Code, OpenClaw, LangChain, or a voice agent.

Let multiple AI agents collaborate through one shared memory — a blackboard where each agent posts findings and reads what the others know, by meaning, in plain SQL. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Build retrieval-augmented generation for any AI agent in one database — embed your docs, retrieve the most relevant chunks by meaning, and generate a grounded answer in plain SQL. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Make an AI agent cite its sources — retrieve passages, generate an answer, and attach the source documents it used, all in one SQL database. Auditable RAG that works with Claude Code, OpenClaw, LangChain, or a voice agent.

Make an AI agent recall the right memories by blending semantic relevance with recency — fresh, on-point context in one SQL query, no separate vector store or cache. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Stop your AI agent from hallucinating — generate an answer, then have the agent verify it against your source data before returning it, all in one SQL database. Self-checking RAG for Claude Code, OpenClaw, LangChain, or a voice agent.

Stop your AI agent from storing the same fact ten times — detect and merge near-duplicate memories by meaning with vector similarity in plain SQL. Keeps agent memory clean for Claude Code, OpenClaw, LangChain, or a voice agent.

Route an AI agent's request to the right tool by meaning — semantic function-calling with vector embeddings in plain SQL, no framework, no brittle keyword rules. Works with Claude Code, OpenClaw, LangChain, or a voice agent.

Train a binary classifier on labelled transactions, deploy it, and score new transactions in real time — the BFSI ground-truth pattern for card-not-present fraud.

Train a binary classifier on logins to catch credential-stuffing + session hijack attempts in real time — the SaaS security ground-truth pattern for account takeover (ATO).

Train a binary classifier on past claims to auto-flag suspect submissions — turn the legacy red-flag-rules engine into a learning model.

Train a classifier to flag patients at high risk of 30-day readmission at discharge time — directly attacks the CMS Hospital Readmissions Reduction Program penalty.

Catch latency-spike anomalies with pure SQL Z-scores — no model training, no infra, runs against any AIDB instance.

Catch silent equipment drift by comparing each sensor reading to that machine's own historical mean/sigma — the Industry-4.0 pattern that beats global thresholds.

Catch HR spikes against each patient's own personal baseline — the clinical pattern that beats fleet-wide thresholds and prevents alert fatigue.

Catch electricity-meter tampering by flagging hours where consumption suddenly drops to near-zero — the utility-fraud pattern that costs the global grid $96B/year.

Use EMBED + COSINE_SIMILARITY to catch claim narratives that semantically match known fraud patterns — the unstructured-data side of SIU fraud detection.

Create a simple table for MP3 audio files with essential metadata for audio file management

Store audio files in multiple formats including MP3, WAV, FLAC, AAC, and OGG for comprehensive audio management

Store podcast episodes with show relationships for building a complete podcast management system

Create a music library with artist and album relationships for comprehensive music management

Store voice memos and recordings with timestamps for personal and business voice note management

Search podcast content via transcribed text for discovering episodes by spoken content

Organize audiobook chapters with sequential ordering for complete audiobook library management

Store customer service call recordings with metadata for compliance and quality assurance

Create playlists with ordered audio tracks for music and podcast playlist management

Catalog sound effects with categories and tags for game development and media production

Store meeting recordings with attendee metadata for corporate meeting management

Store audio files with detected language from transcription for multilingual content management

Create your first AIDB table with basic data types - the essential starting point for learning AIDB SQL

Create a table using all standard SQL types including INTEGER, VARCHAR, BOOLEAN, DECIMAL, and more

Create a table with AUDIO, VIDEO, IMAGE, and PDF columns for storing rich media content

Create a table with VECTOR type for storing embeddings and enabling similarity search

Create tables with PRIMARY KEY, UNIQUE, NOT NULL, and CHECK constraints for data integrity

Create related tables with FOREIGN KEY constraints to establish referential integrity

Create tables with DEFAULT values for automatic population of columns

Create tables with JSON and JSONB columns for flexible metadata and semi-structured data storage

Create tables using UUID for distributed-friendly unique identifiers instead of sequential integers

Create tables with created_at and updated_at timestamp columns for automatic record tracking

Quit demo to show the capabilities of IMMUTABLE tables and security features!

Store and manage PDF documents with metadata for document management systems

Extract and search text content from PDF documents using AI

Process multi-page PDF documents with page-level metadata and content tracking

Use vector embeddings for semantic document search and similarity matching

Extract and store structured data from invoice PDF documents

Store and manage legal documents with contract tracking and compliance features

Track document versions with change history and revision management

Manage contracts with parties, terms, obligations, and renewal tracking

Store and search candidate resumes with skills extraction and matching

Store research papers with citations, authors, and semantic search capabilities

Automatically classify documents into categories using AI and rule-based systems

Extract structured data from PDF forms including fields, values, and validation

Manage document approval workflows with routing, tasks, and audit trails

Build a searchable knowledge base from PDF documents using AI embeddings

Track document compliance requirements with retention policies and audit trails

Manage product images with multiple views, variants, and quality levels for e-commerce

Manage product demonstration videos with timestamps and feature highlights

Manage customer reviews with photos and ratings for social proof

Manage brand assets including logos, guidelines, and marketing materials

Create and manage product comparison visuals for helping customers decide

Manage seasonal promotions, campaign banners, and marketing visuals

Create visual size guides with measurement images and fit recommendations

Manage unboxing videos showing product packaging and first impressions

Organize and manage educational course videos with chapters, transcripts, and progress tracking

Manage university lecture recordings with scheduling, multi-camera capture, and searchable transcripts

Manage multimedia student assignment submissions with versioning and feedback

Manage presentation slides with versioning, annotations, and student notes

Create multimedia quizzes with images, audio, and video questions

Manage online class recordings with breakout rooms, chat logs, and participant tracking

Build a comprehensive multimedia learning resource library with categorization and recommendations

Manage employee training content with completion tracking and certification

Organize marketing assets across campaigns with versioning, approvals, and usage tracking

Archive and search corporate communications including memos, announcements, and recordings

Manage technical product documentation with versioning, localization, and feedback

Manage legal contracts, agreements, and compliance documents with expiration tracking

Manage brand guidelines, logos, and visual identity assets with usage rules

Manage new hire onboarding content including videos, documents, and checklists

Archive and search corporate meeting recordings with transcripts and action items

Build a 6-person social graph and find friends-of-friends with a single Cypher pattern match

Traverse a 4-level reporting hierarchy with variable-length paths to list every direct and indirect report

Recommend movies a user has not seen by traversing watched-movies and shared-genre patterns

Build an order-product-customer graph and surface the top items frequently bought with a target product

Use weighted edges and mutual-friend counts to suggest new connections ranked by tie strength

Detect circular money-laundering patterns up to length 4 by finding cycles in a transaction graph

Trace which products and customers are affected if a critical vendor goes offline

Flag dangerous drug combinations across all patients via a prescribed-drug interaction graph

Resolve what a user can actually do by walking user-group-role-permission graph paths

Topologically order a software dependency graph and find the longest critical path

Rank papers by citation influence using a multi-hop weighted graph traversal

Cluster duplicate customer records that share emails, phones, or addresses across systems

Surface multi-hop transaction chains that obscure source-of-funds, longer than 3 hops with high values

Answer multi-hop natural-language questions by traversing a small entity graph

Combine SIMILAR_TO vector hops with structural patterns to answer questions vector-only RAG can't

Send a paragraph of a CFO call to /v2/graph/extract and query the resulting entity graph

Extract parties, obligations, and dates from contract text, then have LLM_SCORE flag risky clauses

Recommend "products like this" by walking SIMILAR_TO over embedding properties on Product nodes

Build a long-term episodic memory graph for an LLM agent, recall by semantic + structural patterns

Match candidates to roles with structural skill hits plus semantic similarity and an LLM fit score

Catch duplicate customers whose emails differ but whose name + address embed alike

Cluster IOCs and actors across feeds by extracting from advisories and matching with SIMILAR_TO

Extract supplier-event edges from news prose and grade each event with LLM_SCORE

Walk drug-target-disease chains then use SIMILAR_TO on mechanism embeddings to find candidates

Extract entities from abstracts and cluster papers semantically with SIMILAR_TO

Combine connection-count centrality with LLM_SCORE for authenticity, surface real influencers

Walk transitive dependency edges and have the LLM judge if a CVE actually applies in your context

Use SIMILAR_TO over symptom embeddings on patient nodes to surface clinical lookalikes

Extract action items, owners, and dates from meeting minutes and rank urgency with LLM_SCORE

Store and manage medical images with patient records and metadata

Organize and manage radiographic X-ray images with views and annotations

Manage complex cross-sectional imaging studies with multiple sequences and reconstructions

Manage whole slide images for digital pathology and tissue analysis

Create and manage annotations, measurements, and findings on medical images

Track and query patient imaging studies over time for longitudinal analysis

Link radiology reports with images and extract structured findings

Analyze medical imaging data for operational insights and quality metrics

Store images in multiple formats including JPEG, PNG, WEBP, and GIF with format-specific columns

Store images with dimensions, file size, and EXIF data in JSON columns for rich metadata management

Create albums and photos tables with relationships for organizing images into collections

Implement a flexible tagging system for images using a junction table for many-to-many relationships

Store user profile pictures with thumbnail variants for efficient display at different sizes

Extract text from images using the EXTRACT_TEXT function for searchable image content

Query images by dimensions using the DIMENSIONS function for filtering by size and aspect ratio

Create embeddings for images to enable similarity search and visual content discovery

Store and organize application screenshots with metadata for bug tracking and documentation

Store multiple image sizes for e-commerce products including thumbnails, medium, and full-size variants

Store scanned documents as images with OCR text extraction for searchable document archives

Train a simple regression model, deploy it, run predictions, then verify results match the known function.

Build a model to predict which customers are likely to churn based on their account characteristics and usage patterns.

Score compounds for likelihood of success. Prioritize top candidates for lab testing.

Build an intelligent real estate platform that matches buyers with properties, predicts market trends, analyzes property values, and generates insi...

Create intelligent marketing campaigns by analyzing customer behavior, segmenting audiences, predicting churn, and generating personalized content...

Store and manage surveillance video footage from security cameras

Track access control events with badge photos and entry images

Store and query license plate recognition data from parking and security systems

Log and query facial recognition events for access control and security

Manage video evidence collection and chain of custody for security incidents

Manage security alerts, notifications, and response workflows

Store and retrieve MP4 video files with metadata for video content management

Store videos in multiple formats (MP4, AVI, MKV, WEBM, MOV) for cross-platform compatibility

Query and analyze video content by duration for content planning and scheduling

Extract and store video frames for thumbnails, previews, and image analysis

Extract and manage audio tracks from video files for transcription and analysis

Transcribe video content to text for searchability, subtitles, and accessibility

Add chapter markers to videos for enhanced navigation and content organization

Store multiple thumbnail images for video content with selection preferences

Organize videos into hierarchical categories with tags for content discovery

Manage video series with episodes, seasons, and sequential ordering

Build a platform for user-uploaded videos with moderation and engagement features

Track video views with detailed analytics including watch time and completion rates

Store multi-language subtitles and captions for video accessibility

Build a complete voice support agent on one database — caller memory, real-time skill routing, RAG-grounded spoken answers, an urgency model, and a semantic cache, all in SQL. The brain for any STT/TTS stack (Vapi, LiveKit, Twilio, Pipecat).

Make a voice AI agent recognize returning callers — recall their past calls and preferences by meaning and map their account relationships as a graph, so it greets them by context, not a cold script. For any STT/TTS stack (Vapi, LiveKit, Twilio, Pipecat).

Give a voice AI agent memory across a call and between calls — store transcribed turns, recall by meaning, and keep a rolling summary so it never loses the thread. Database-as-the-brain for any STT/TTS stack (Vapi, LiveKit, Twilio, Pipecat).

Cut voice AI latency and cost with a semantic answer cache — serve a stored answer when a caller's question means the same as one you've answered before, even if worded differently, with a vector lookup in SQL. For any STT/TTS stack (Vapi, LiveKit, Twilio, Pipecat).

Route a voice AI agent's caller to the right skill in real time — match the transcribed utterance to a skill by meaning with a vector lookup, fast enough for a live call. Database-as-the-brain for any STT/TTS stack (Vapi, LiveKit, Twilio, Pipecat).

Ground a voice AI agent's spoken answers in your own knowledge base — retrieve the right passage by meaning and generate a short, speakable, accurate reply, so the agent never makes things up on a call. For any STT/TTS stack (Vapi, LiveKit, Twilio, Pipecat).