AI Apps Directory

Open-source RAG engine with deep document understanding.

RAGFlow is an open-source retrieval-augmented generation engine that turns complex documents—PDFs, slides, spreadsheets, scans, and web pages—into grounded, citation-backed context for LLMs. Its DeepDoc parser and hybrid vector plus full-text search aim for reliable, hallucination-resistant question answering, and it now bundles an agent-orchestration layer. Self-host the Apache-2.0 engine or use the managed cloud.

Natural-language data analyst that turns questions into SQL.

Defog is an AI data analyst that converts natural-language questions into SQL across databases, warehouses, and CSVs, returning answers without writing queries. It is built on SQLCoder, the company's own open-source text-to-SQL model, and emphasizes privacy—your data is never sent to or used to train the model. Defog Agents extend it to multi-step SQL, Python, and R workflows.

Zero-shot AI image annotation that batch-labels with visual prompts.

T-Rex Label is a browser-based image annotation tool built on the T-Rex2 open-set detection model. Point it at one example and its visual-prompt, zero-shot detection finds and labels matching objects across an entire dataset—no training or fine-tuning required—handling dense, occluded, and varied-lighting scenes. It exports to COCO and YOLO formats and integrates with tools like Roboflow and Labelbox.

Real-time generative audio API for music, sound effects, and speech.

CassetteAI is a generative audio platform that produces music, sound effects, and speech from text prompts through a single API. Its latent-diffusion models render a 30-second sample in roughly two seconds and full multi-minute tracks at 44.1 kHz stereo, aimed at creators and developers who need audio on demand. It is offered pay-per-use with a free monthly tier.

Fully managed vector database service built on open-source Milvus.

Zilliz Cloud is a fully managed vector database from the creators of Milvus. It runs billion-scale similarity search for RAG, semantic search, and AI applications without the operational burden of self-hosting, offering a free serverless tier, auto-scaling serverless, and dedicated clusters across major clouds. Features include GPU-accelerated indexing, tiered storage, and hybrid dense/sparse search.

Open-source infrastructure for computer-use agents across full desktops.

Cua (also styled c/ua) is infrastructure for building and scaling computer-use agents. One API boots Linux, Windows, macOS, and Android machines in lightweight, isolated virtual containers, and SDKs plus benchmarks let agents observe and control full operating systems — clicking, typing, browsing, and running real apps. On Apple Silicon it reaches near-native speed via Apple's Virtualization.framework and the Lume CLI, and it runs locally with any LLM or on Cua's managed cloud.

Purpose-built AI models and infrastructure for coding agents.

Relace builds specialized models and infrastructure that slot into AI code-generation products. Its Instant Apply model merges partial diffs from frontier models into full files at thousands of tokens per second, and its two-stage code retrieval (embedding search plus a code reranker) finds the right context fast. Relace also offers managed repository hosting with automatic per-commit indexing, so coding agents get cheaper, faster, more reliable edits and search.

Open-source inspector to test, debug, and evaluate MCP servers and ChatGPT apps.

MCPJam is a local-first developer tool, a 'Postman for MCP', for testing, debugging, inspecting, and evaluating Model Context Protocol servers. You can manually run tools, resources, resource templates, prompts, and elicitation flows with full JSON-RPC observability, debug every message and OAuth exchange, and run evaluations across multiple LLMs to catch regressions. It runs as a hosted web app, a Mac/Windows desktop app, or via npx in your terminal.

Personalized, real-time soundscapes that adapt to help you focus, relax, and sleep.

Endel generates personalized, adaptive soundscapes in real time, blending music and atmospheric elements with techniques like binaural beats and colored noise. Its patented Endel Pacific engine responds to inputs such as time of day, weather, location, circadian rhythm, and heart rate to create sound for focus, relaxation, sleep, and activity. The app spans a wide device ecosystem, from phones and Mac to Apple Watch, Apple TV, and in-car wellness systems.

An autonomous AI SRE that investigates production alerts and finds root cause.

Cleric is an AI agent for site reliability engineering that automates incident response. When an alert fires, it investigates across your observability dashboards, performs root-cause analysis in minutes, recommends and verifies fixes against the live environment, and retains what it learns as institutional knowledge for the team. It works inside Slack and existing tooling with read-only access by default, adding write access only when teams are ready.

Build, curate, and host Model Context Protocol servers from your APIs or TypeScript.

Gram is a platform for creating, curating, and hosting Model Context Protocol (MCP) servers. Instead of exposing raw API endpoints that confuse agents, it lets you compose higher-order, business-level tools from an OpenAPI document or TypeScript functions, group them into toolsets, and deploy each as a hosted, authenticated MCP server. Every toolset is instantly usable from Claude, ChatGPT, Cursor, or any MCP-compatible client, with an optional OAuth 2.1 proxy and org-wide access controls.

Security posture and threat detection for SaaS and AI agents.

Reco is a SaaS security platform that discovers and secures the apps, identities, and AI agents running across an organization. It maps human and non-human identities, surfaces shadow AI and misconfigurations, and detects threats across 200+ integrated applications. Its AI Agent Security adds visibility and control over autonomous agents like Copilot and Agentforce.

AI agents that investigate financial fraud for institutions.

Finic builds AI agents that automate fraud investigations for banks and fintechs. Its agents connect to existing tools through the browser like a human analyst, pull structured and unstructured data, triage cases, and write up case notes — letting fraud-ops teams clear far more reviews without added headcount. Founded by engineers from Robinhood's identity team.

Open-source text- and image-to-video generation from Alibaba.

Wan is an open-source family of video generation models from Alibaba's Tongyi Lab, covering text-to-video, image-to-video, and image generation and editing. Released under Apache 2.0 with weights, training, and inference code public, it can be self-hosted or used free on the wan.video site. Successive versions, including Wan 2.2's mixture-of-experts architecture, have topped open-video benchmarks.

Transcribe, subtitle, dub, and voiceover in 125+ languages.

Maestra is a cloud media platform that converts audio and video to text, then generates translated subtitles, AI voiceovers, and dubbing across 125+ languages. It handles transcription, captioning, translation, and live captions in one workspace, with an editor for cleanup and export. Built by Katara Tech for creators and enterprise localization teams.

AI coding agent for your IDE and terminal.

Zencoder is an AI coding agent that indexes an entire repository to understand its architecture and dependencies, then writes, tests, and fixes code across the development lifecycle. Its Zen Agents are customizable, shareable autonomous agents for tasks like generating tests, updating dependencies, or opening PRs, while Zentester focuses on automated unit and end-to-end tests. It runs inside popular IDEs like VS Code and JetBrains and via a CLI, supporting 70+ languages.

Speech-to-speech platform for reliable voice agents.

Phonic is a platform for building production voice agents on its own end-to-end speech-to-speech models, rather than chaining separate speech-to-text, LLM, and text-to-speech stages. It targets sub-300ms latency for natural turn-taking and reliable tool calling, and bundles evaluation, session records, and real-time observability to surface failure points. Aimed at enterprises, it offers cloud API access plus containerized deployment in your own environment.

Rerankers and embeddings that sharpen AI retrieval.

ZeroEntropy builds specialized models — rerankers, embeddings, and custom retrieval models — that improve search accuracy in RAG pipelines and agentic apps. Its zerank rerankers reorder retrieved results so the most relevant passages reach the model, reportedly beating Cohere and Gemini rerankers at lower cost. The models are served through a single, latency-optimized API with Python and TypeScript SDKs, and the smaller zerank model is released as open weights.

Sandbox-native cloud for AI agents.

Tensorlake is a cloud platform that gives AI agents their own isolated, stateful sandboxes for running code and calling tools safely. Each sandbox is a microVM that can pause and resume, so long-running agentic loops keep their state across restarts and can run for hours. It exposes serverless Python workflow APIs that scale to zero when idle, and its earlier document-parsing and OCR API now runs as one workload on top. A free tier is available.

Explore, structure, and analyze millions of unstructured records as interactive embedding maps.

Nomic Atlas is a data-intelligence platform that embeds large collections of text, image, and audio data and renders them as interactive 2-D maps you can browse, search, cluster, and label in the browser. Powered by Nomic's own embedding and topic-modeling models, it scales from hundreds to tens of millions of points and exposes the same pipeline through a developer API for embeddings and retrieval.

High-accuracy document parsing — PDFs and images to markdown, JSON, and HTML.

Datalab turns PDFs, images, and office documents into clean markdown, JSON, and HTML with layout, table, math, and code preservation. It is the commercial, hosted layer over the open-source Marker converter and Surya OCR toolkit, offered as a pay-as-you-go API with a free monthly allowance, while the underlying models stay free to self-host for research and small startups.

AI-powered studio for recording, editing, and producing audio and video.

Podcastle is a browser-based content-creation platform for podcasters and video creators: multi-track remote recording, AI text-to-speech and voice cloning, transcription, and one-click audio and video editing with noise removal and leveling. It bundles studio capture, AI voices, and editing into a single workflow so creators don't need a separate DAW or video editor.

Best-in-class embedding and reranking models for retrieval and RAG.

Voyage AI builds retrieval-specialized embedding and reranking models served via API to ground LLM applications. Its voyage-3 series of text and code embeddings, domain-tuned variants, and rerank models (e.g. rerank-2.5) are aimed at higher RAG accuracy than general-purpose embeddings. Now part of MongoDB, its models are being woven into Atlas Vector Search while the standalone API continues to operate.

Multi-model AI art generator wrapped in a creative community.

NightCafe is a browser-based AI art platform that generates and edits images from text prompts across many engines in one place — FLUX, Stable Diffusion, DALL·E, Google Imagen, Ideogram and more — plus image-to-video and style-transfer tools. It runs on daily free credits and a credit-pack/PRO model, and is built around an active community with shared galleries, chat rooms, and a daily AI Art Challenge.