The models, editors, frameworks, stores and platforms that make up a working AI stack — grouped by the job they do rather than by who sells them. Each entry says what the tool is genuinely good at and where it stops being the right answer.
Nothing here is ranked or rated. There are no scores, no user counts and no paid placement, and pricing is recorded as a commercial model rather than a figure we could not keep accurate.
32 of 32 tools
Unified model interface for JavaScript applications
The AI SDK gives JavaScript and TypeScript applications a single interface across model providers, covering text generation, structured output, tool calling and embeddings. Its streaming…
Best forWeb applications that need streaming AI interfaces today and the freedom to swap model providers later.
Agentic coding in the terminal
Claude Code is Anthropic’s coding agent that runs as a terminal process and operates directly on a repository — it reads files, edits them, runs commands and iterates on the output it gets…
Best forMulti-file changes where the agent needs to read the surrounding code, run the test suite and iterate on real failure output.
AI-native code editor
Cursor is a fork of VS Code rebuilt around model-assisted editing: inline completion, a chat pane with codebase context, and an agent mode that applies edits across several files. It…
Best forDevelopers who want agentic edits and repo-aware chat without leaving a familiar VS Code environment.
Completions and chat inside every editor
GitHub Copilot provides inline completion, chat and agent-style edits across the major editors, backed by a selectable model. It draws context from open files and the surrounding project,…
Best forOrganisations already on GitHub that want one governed assistant available in every editor their teams use.
Open-source LLM tracing and evaluation
Langfuse captures traces of LLM applications as nested spans — one per model call, retrieval step and tool invocation — with inputs, outputs, latency and token counts attached. On top of…
Best forTeams that need production tracing and evaluation while keeping trace data inside their own infrastructure.
Graph-structured agent runtime
LangGraph models an agent as a state graph: nodes are steps, edges are transitions, and a shared state object flows between them. Because control flow is explicit rather than implicit in a…
Best forLong-running or approval-gated agent workflows that need durable state and control flow you can read off a diagram.
Run open-weight models locally
Ollama packages open-weight models as pullable artefacts and serves them over a local HTTP API, handling quantised weight downloads, GPU or CPU placement and model lifecycle. A Modelfile…
Best forLocal development, offline work and privacy-sensitive workloads where the data must not leave the machine.
Managed vector database
Pinecone is a fully managed vector database that hides index sharding, replication and scaling behind an upsert-and-query API. Metadata filtering runs alongside similarity search, and…
Best forTeams that want production vector search without owning the storage layer or its failure modes.
Evaluation and experiment tracking for AI
Braintrust is built around evaluation: you define datasets, scoring functions and experiments, then compare runs as prompts, models or application code change. Production logs feed back…
Best forTeams that want prompt and model changes gated by a measurable score rather than by how the output felt.
Anthropic’s model API for long-context and agentic work
The Claude API serves Anthropic’s model family with tool use, structured output, long-context requests, prompt caching and batch processing. Extended thinking lets you trade latency for…
Best forLong-context reasoning, careful instruction following and agentic tool use with explicit control over cost and latency.
Open-source assistant you configure yourself
Continue is an open-source IDE extension for VS Code and JetBrains offering completion, chat and edit modes against whatever model you point it at — a hosted API or a local runtime. Its…
Best forTeams that need an assistant running against self-hosted or security-approved models under configuration they control.
Role-based multi-agent orchestration
CrewAI structures multi-agent systems around roles, goals and tasks: you declare agents with a purpose and a toolset, then compose them into a crew that executes sequentially or…
Best forWorkflows that decompose cleanly into distinct specialist roles with a clear hand-off order between them.
Model, dataset and inference hub
Hugging Face hosts open model weights, datasets and demo applications on a Git-backed hub that versions large artefacts through LFS. Its libraries — transformers, datasets, accelerate and…
Best forFinding, evaluating and fine-tuning open-weight models without building your own distribution infrastructure.
Event-driven durable functions
Inngest turns ordinary functions into durable, event-triggered steps: each step boundary is checkpointed, so a retry resumes from the last completed step instead of the beginning. Flow…
Best forAdding retries, concurrency limits and long waits to AI workflows that already live inside a serverless application.
Tracing and evals with deep LangChain support
LangSmith records traces, prompts and token usage for LLM applications, with first-class support for LangChain and LangGraph runs and a framework-agnostic SDK underneath. It pairs the…
Best forDebugging and evaluating applications already built on the LangChain or LangGraph stack.
Data framework for retrieval over your own content
LlamaIndex concentrates on the ingestion half of an LLM application: loading documents, chunking them, building indexes, and querying through retrievers and query engines. The storage…
Best forRetrieval pipelines over heterogeneous document sources where parsing, chunking and indexing are the hard part.
Serverless GPU compute defined in Python
Modal runs Python functions on remote CPU or GPU containers described in code: the image, hardware and scaling policy are decorators on the function rather than separate infrastructure…
Best forBursty GPU workloads where paying for idle capacity is precisely the cost you are trying to avoid.
Workflow automation with AI nodes
n8n is a node-based workflow automation tool with a visual editor, a large connector library and an escape hatch into JavaScript when no node fits. Its AI nodes let a workflow call a…
Best forWiring an AI step into the operational systems around it — email, CRM, storage, webhooks — without writing an integration for each.
React framework with server rendering built in
Next.js is a React framework whose App Router makes Server Components the default, so data fetching and rendering happen on the server and only interactive leaves ship JavaScript.…
Best forContent and product sites that need server rendering, SEO control and streaming UI from a single codebase.
Hosted model APIs across text, image and audio
OpenAI’s platform exposes its model family through a REST API with tool calling, structured output, embeddings, audio and image generation. Alongside inference it offers batch processing,…
Best forTeams that want broad modality coverage and the widest library and framework support behind a single API key.
One endpoint in front of many model providers
OpenRouter is a routing layer that exposes models from many vendors behind a single OpenAI-compatible endpoint, with unified billing and per-request routing rules. Provider fall-back means…
Best forComparing or hot-swapping models across vendors without maintaining a separate integration for each one.
Vector similarity search inside PostgreSQL
pgvector is a PostgreSQL extension adding vector column types and distance operators, with HNSW and IVFFlat index types for approximate nearest-neighbour search. Because vectors live in…
Best forApplications already on PostgreSQL that want retrieval without adding another datastore to operate and keep in sync.
Typed agents with validated outputs
Pydantic AI applies Pydantic’s validation model to agent construction: tool arguments and final results are declared as schemas, and the framework validates the model’s output against…
Best forPython services where an agent’s output has to satisfy a schema before it is allowed to touch anything else.
Vector search engine with first-class filtering
Qdrant is a vector search engine written in Rust that pairs HNSW indexing with a filterable payload store, applying metadata conditions during the search rather than discarding results…
Best forFiltered vector search at scale, where predicate performance and memory cost are the constraints that bite first.
Run and fine-tune models behind an API
Replicate hosts open and proprietary models as versioned, API-callable endpoints, with a container packaging format that defines each model’s environment and input schema. Predictions are…
Best forAdding image, audio or specialised open models to a product without standing up GPU infrastructure.
Postgres backend with auth, storage and edge functions
Supabase assembles a managed PostgreSQL database with generated REST and realtime APIs, authentication, object storage and edge functions, keeping SQL and row-level security as the…
Best forApplications that want a relational database plus the surrounding services without giving up SQL or portability.
Durable execution for long-running workflows
Temporal runs workflows as durable code: each step’s result is persisted to an event history, so a process survives crashes, deploys and multi-day waits by replaying deterministically from…
Best forMulti-step agent or business workflows that must survive process restarts and stay alive for hours or days.
Deployment platform for frontend and AI applications
Vercel deploys frontend frameworks and serverless functions with per-branch preview environments, an edge network and build-time framework integrations. Streaming responses, background…
Best forShipping frontend and AI-facing applications with preview deployments and close to no infrastructure work.
High-throughput inference server for open models
vLLM is an inference engine whose PagedAttention memory manager treats the key-value cache like virtual memory pages, letting it hold many concurrent sequences without fragmenting GPU…
Best forSelf-hosting open-weight models where concurrent throughput per GPU is the number you are optimising.
Open-source vector database with hybrid search
Weaviate stores objects together with their vectors and can generate the embeddings itself through pluggable vectoriser modules at write time. Dense vector search and keyword BM25 scoring…
Best forHybrid semantic and keyword retrieval where you want the database to own embedding generation rather than your pipeline.
Agentic IDE with a persistent task context
Windsurf is an AI-first IDE whose agent can read the project, plan an edit spanning several files and run terminal commands while keeping a running model of what it has already changed. It…
Best forLonger, multi-step tasks where you want the agent to keep working while you review its diffs.
High-performance editor with a bring-your-own-model assistant
Zed is a GPU-accelerated editor written in Rust, designed around low input latency and real-time multiplayer editing. Its assistant panel talks to whichever model provider you configure,…
Best forDevelopers who care about editor latency and want AI assistance without a heavyweight Electron shell.
These rules are why the directory is worth reading rather than being another affiliate list. They apply to every entry, and they are the same rules whether or not a vendor has ever heard of us.
Every entry is a product you can open in a browser today. Each one is described by what it actually does — the mechanism, the integration surface, the workflow it assumes — rather than by its marketing copy.
There is no star rating, no review count and no benchmark figure in this directory, because none of them could be verified or kept current. Pricing is recorded as a commercial model only — free, freemium, paid or open source — never as a figure.
Each tool records the situation where it is the wrong answer: the constraint it imposes, the workload it handles badly, or the lock-in it creates. An entry without an honest limitation would not be an assessment.
Ordering is featured-then-alphabetical and nothing about it can be bought. Outbound links point at each product’s root domain with no referral, affiliate or campaign parameters attached.
Tooling moves quickly, and an entry describes a product as we understand it at the time of writing. Where a description no longer matches the product, the entry is wrong and we would rather hear about it — tell us what changed.
Head-to-head breakdowns of the choices these tools force on you — architecture and trade-offs, never scores.
Plain definitions for the vocabulary the directory uses: retrieval, tool calling, quantisation, evaluation.
Longer, criteria-first assessments for the tools that deserve more than a directory entry.