AI Developer Tools for LLM APIs and GPU Clouds
Tools that call the LLM APIs or rent the GPUs we track, and which providers each one works with. Tools are listed by name within each category.
AI coding agents
Compare AI coding agents →Agents and editor extensions that read, write and run code by calling an LLM API.
Open-source pair-programming tool that runs in the terminal and edits code in a local git repository. It connects to hosted LLM APIs or local models.
Open sourceSelf-hostableFree tierWorks with 10 of the LLM providers we trackAnthropic's agentic coding tool. It reads a codebase, edits files and runs commands from the terminal, IDE extensions, a desktop app or the browser, using Claude models through an Anthropic account or a supported cloud provider.
Works with 6 of the LLM providers we trackOpen-source coding agent available as an IDE extension, a CLI and an SDK. It plans and makes code changes and runs commands, using a model provider the user selects or Cline's own pay-as-you-go credits.
Open sourceSelf-hostableFree tierWorks with 12 of the LLM providers we trackOpen-source coding agent for VS Code, JetBrains IDEs and the command line. Users can bring their own API keys, run local models, or route requests through Kilo's own gateway.
Open sourceSelf-hostableFree tierWorks with 12 of the LLM providers we trackOpen-source coding agent that runs in the terminal. It connects to model providers through a /connect command or a config file, and supports local runtimes such as Ollama, LM Studio and vLLM.
Open sourceSelf-hostableFree tierWorks with 12 of the LLM providers we track
AI chat apps
Compare AI chat apps →Desktop, web and self-hosted chat interfaces that connect to one or more LLM APIs.
Desktop and Docker application from Mintplex Labs for chatting with documents (retrieval-augmented generation) and running agents. It uses either a built-in local model or a connected LLM provider. The Docker version adds multi-user workspaces and permissions, and the vendor also offers hosted instances.
Open sourceSelf-hostableFree tierWorks with 12 of the LLM providers we trackOpen-source desktop chat application from Menlo Research. It downloads and runs open-weight language models locally and offline, and can also connect to cloud model APIs with the user's own API keys.
Open sourceSelf-hostableFree tierWorks with 8 of the LLM providers we trackOpen-source, self-hosted chat interface that connects to many AI model providers from one application. It includes agents, MCP tool connections, a code interpreter and multi-user authentication. Built-in endpoints and custom OpenAI-compatible endpoints are configured in a librechat.yaml file.
Open sourceSelf-hostableFree tierWorks with 12 of the LLM providers we trackSelf-hosted web interface for chatting with language models. It connects to local runtimes such as Ollama and to OpenAI-compatible APIs, and runs on the user's own server or machine (for example with Docker), including fully offline.
Self-hostableFree tierWorks with 9 of the LLM providers we track
LLM gateways and routers
Compare LLM gateways and routers →Proxies that put many LLM APIs behind one endpoint, with routing, fallbacks, caching and spend limits.
Hosted proxy on Cloudflare's network between an application and AI model providers. It adds analytics, logging, caching, rate limiting, retries, fallbacks and spend limits, reached through provider-specific endpoints or an OpenAI-compatible unified endpoint.
Free tierWorks with 15 of the LLM providers we trackOpen-source AI gateway and LLM observability platform. Applications route requests through its OpenAI-compatible gateway or provider-specific proxy URLs, and Helicone logs requests, cost and latency and handles provider routing and fallbacks.
Open sourceSelf-hostableFree tierWorks with 15 of the LLM providers we trackOpen-source Python SDK and self-hosted proxy server that lets applications call many LLM providers through one OpenAI-compatible API. Teams run the proxy as a shared gateway with virtual keys, budgets, spend tracking, routing, fallbacks and guardrails.
Open sourceSelf-hostableFree tierWorks with 15 of the LLM providers we trackAI gateway that routes requests to many LLM providers through one API, with retries, fallbacks, load balancing, conditional routing and guardrails. The gateway is open source and can be self-hosted; the hosted platform adds logging, observability and prompt management.
Open sourceSelf-hostableFree tierWorks with 15 of the LLM providers we track
LLM observability and evals
Compare LLM observability and evals →Tracing, monitoring and evaluation of the calls an application makes to LLM APIs.
Open-source platform for tracing, evaluating and debugging LLM applications and agents, with prompt management and datasets. It is available as a hosted service (Langfuse Cloud) or can be self-hosted with Docker Compose or Kubernetes.
Open sourceSelf-hostableFree tierWorks with 15 of the LLM providers we trackLangChain's platform for tracing, monitoring and evaluating LLM applications and agents, usable with or without the LangChain framework through its Python and TypeScript SDKs. It is a hosted service; self-hosting is an add-on to the Enterprise plan.
Self-hostableFree tierWorks with 6 of the LLM providers we track
GPU orchestration
Compare GPU orchestration →Tools that launch, schedule and manage GPU workloads across clouds and clusters.
Open-source control plane for GPU provisioning and orchestration that runs dev environments, training tasks and inference services across GPU clouds, Kubernetes, Slurm and on-prem servers. Configurations are YAML files applied with the dstack CLI against a self-hosted server or the hosted dstack Sky.
Open sourceSelf-hostableFree tierWorks with 15 of the GPU clouds we trackOpen-source system for launching and managing AI jobs, clusters and model serving across Kubernetes, Slurm and cloud providers from one CLI. A separate commercial SkyPilot Platform adds managed hosting and team features such as quotas, SSO and support.
Open sourceSelf-hostableFree tierWorks with 15 of the GPU clouds we track
LLM fine-tuning
Frameworks for fine-tuning open-weight models on rented or owned GPUs.
Open-source framework for fine-tuning large language models (full fine-tuning, LoRA and QLoRA, RLHF and GRPO) from a YAML configuration, with multi-GPU and multi-node training. It is distributed as a PyPI package and Docker images for local hardware and cloud GPU providers.
Open sourceSelf-hostableFree tierWorks with 10 of the GPU clouds we track
Experiment tracking
Logging, comparing and versioning training runs, models and LLM traces.
Open-source platform for tracking machine learning experiments (parameters, metrics, artifacts), versioning models in a model registry, and tracing and evaluating LLM and agent applications. Teams run its tracking server themselves or use a managed MLflow service.
Open sourceSelf-hostableFree tierWorks with 13 of the LLM providers we track
