AI & LLM Tools

Gemini Pro / Flash / Nano with huge context windows and native multimodal input. Free tier is generous.

Grok models via Twitter-adjacent API. Competitive reasoning and real-time web grounding.

Open-weight and hosted models out of Paris. Strong European option with EU data residency.

Command R / R+ models optimized for RAG and enterprise workloads. Strong rerank + embed endpoints.

Hosts 100+ open models with OpenAI-compatible APIs. Great for switching providers without rewriting code.

Fast serverless inference for open-weight models with fine-tuning support.

Run any open-source model behind an HTTP API. Great for vision / audio / niche models.

Pay-per-token OpenAI-compatible API for open models. Aggressive pricing.

Observability + guardrails + routing layer for production LLM apps. 50+ providers.

Open-source LLM observability. Proxies requests and captures logs, cost, latency.

Desktop app for chatting with local models. Great for non-technical users.

The CPU/GPU inference engine that started the local-LLM revolution. Powers most of the stack.

High-throughput batched inference for serving open-weight models in production.

State-machine-style agent framework from the LangChain team. Durable workflows + checkpointing.

Data framework optimized for RAG — indexing, retrieval, query engines. TS + Python.

TypeScript-first agent framework with memory, tools, workflows, and eval built in.

Multi-agent framework organized around role-based crews. Popular in the no-code adjacent space.

Drag-and-drop UI for LangChain flows. Self-hostable, open-source, great for prototyping agents.

Visual authoring for LangChain pipelines. Big community, plays well with LangGraph.

The canonical no-code automation platform. AI actions across 7000+ apps.

Polished WebUI built for Ollama + any OpenAI-compatible API. One of the cleanest self-hosted options.

Desktop + Docker app that turns documents into a private RAG-powered workspace.

Polished chat UI that lives in your browser, keys stay local. Supports every major provider.

Text-to-video with PikaSwaps, PikaScenes, and lip-sync. Strong on stylized, social-friendly clips.

Ray2 model with fast generation, image-to-video, and keyframe control. API available for builders.

Kuaishou's video model — strong physics, long durations, and competitive prompt adherence. Popular for cinematic shots.

MiniMax's video gen. Standout for realistic motion and human subjects. API via the MiniMax platform.

AI avatars for talking-head videos — realistic lip sync from a script in any language. Used heavily for product demos and marketing.

Enterprise-grade AI video presenters. 230+ avatars, 140+ languages, custom-avatar pipeline. Locked-down compliance story.

Turn a single photo into a talking avatar. Real-time conversational agents and historical-figure recreations.

Genmo's open-source 10B video diffusion model under Apache 2.0. Solid baseline for self-hosted research and fine-tuning.

Speech-to-text plus audio intelligence — diarization, sentiment, summarization, PII redaction. Great DX.

Sonic TTS at sub-100ms latency. The pick for real-time voice agents that need to feel alive, not robotic.

EVI — empathic voice interface that reads tone and adapts. Useful where pitch and pacing matter as much as words.

Ultra-low-latency TTS optimized for voice agents. 600+ voices, 100+ languages, instant voice cloning.

Voice cloning, real-time dubbing, and an audio-watermark / deepfake-detection toolkit. Enterprise focus.

STT with broad accent and dialect coverage. Strong on hard audio (meetings, accents, noisy mics).

Text-to-music in the same league as Suno, with stem separation and longer compositions.

Stability AI's text-to-audio model — instrumentals, sound effects, and stems. Open-weights checkpoints available.

LangChain's commercial observability + eval platform. Tight LangChain / LangGraph integration.

Production-grade eval platform for shipping LLM features safely. Dataset + experiment tracking.

CLI + YAML for running prompt / red-team tests. Great in CI.

Open-source library for tracing + scoring LLM apps. RAG triad metrics and feedback functions out of the box.

OpenAI's official eval framework + registry of public benchmarks. Reference implementation for model-graded evals.

UK AI Safety Institute's eval framework — capability, agent, and safety evals with sandboxed tool use.

Open-source observability + eval for LLMs and agents. Traces, datasets, and prompt experiments in one tool.

EleutherAI's harness — the de facto standard for academic benchmarks (MMLU, GSM8K, HellaSwag, ARC, etc.).

Hugging Face's standardized benchmark leaderboard for open-weight models. Run on Eleuther's harness.

Contamination-resistant benchmark with monthly-refreshed questions. Maintained by Abacus AI + NYU + Meta.

Stanford CRFM's Holistic Evaluation of Language Models — multi-metric scenarios across many models.

Coding-focused leaderboards: polyglot benchmark across six languages, plus refactor + edit-format scoring.

Side-by-side comparison of frontier models on quality, latency, cost, and context window. Updated frequently.

Managed vector DB with the most mature enterprise feature set. Serverless tier is cheap.

Open-source Rust vector DB. Self-hostable or managed cloud. Great filtering story.

Open-source vector DB with hybrid search and module-based embedders. Good GraphQL API.

Serverless vector + full-text search built on object storage. Cheap at scale.