AI/ML Infrastructure & LLM Tools brand directory

Braintrust is a unified AI observability and evaluation platform that helps engineering and product teams trace LLM production traffic, run structured evals, manage and version prompts, and catch regressions before they reach users—powered by Brainstore, a purpose-built database for AI trace data, and Loop, an AI agent for autonomous eval optimization.

LangChain provides a full-lifecycle agent engineering platform combining open-source frameworks (LangChain for rapid agent prototyping, LangGraph for stateful graph-based agent orchestration, Deep Agents for long-running autonomous tasks) with LangSmith, a commercial SaaS product offering trace-level observability, automated and human-in-the-loop evaluation, managed agent deployment with memory and durable checkpointing, and Fleet—a natural-language no-code agent builder for business users. The platform supports Python, TypeScript, Go, and Java SDKs, native OpenTelemetry, MCP and A2A protocol integration, and over 100 integrations with LLM providers and vector databases.

Weights & Biases is an end-to-end AI developer platform spanning ML model development (experiment tracking, hyperparameter sweeps, artifact versioning, model registry) and LLM/GenAI application development (tracing, evaluation, guardrails, agent monitoring via W&B Weave), plus serverless LLM fine-tuning and hosted open-source model inference. Now a subsidiary of CoreWeave.

MLflow is the leading open-source, Apache 2.0-licensed AI engineering platform covering the complete lifecycle of ML models, LLM applications, and AI agents. Its core modules—experiment tracking, model registry, LLM tracing (built on OpenTelemetry), GenAI evaluation, prompt management, AI gateway, and agent deployment server—are available as a unified self-hosted platform or as a managed service via Databricks, AWS SageMaker, and Azure ML. It integrates with 100+ frameworks and supports Python, TypeScript/JavaScript, Java, and R.

Langfuse is an open-source LLM engineering platform that covers the full AI application development lifecycle: hierarchical tracing and agent observability (OTel-native), prompt management with versioning and caching, automated and human evaluation pipelines, structured experiments, and production cost/latency dashboards. It is framework- and model-agnostic, self-hostable under MIT license, and integrates with 80+ tools including LangChain, LiteLLM, LlamaIndex, OpenAI, and Anthropic. Following its January 2026 acquisition by ClickHouse, its ClickHouse-backed data layer supports billions of monthly observations at enterprise scale.

Modal is a serverless AI infrastructure platform that transforms any Python function into an autoscaling cloud workload through a decorator-based SDK requiring no YAML, Dockerfiles, or Kubernetes configuration. Its core products include: Modal Inference (LLM and generative model serving with sub-second cold starts), Modal Training (single- and multi-node GPU fine-tuning), Modal Sandboxes (ephemeral, isolated containers for running AI-generated or untrusted code), Modal Batch (massively parallel CPU/GPU batch jobs), and Modal Notebooks (GPU-backed collaborative notebooks with memory snapshots). The platform is built on Modal's own custom container runtime, filesystem, scheduler, and image builder, pooling capacity across multiple clouds to provide elastic GPU access without quotas or reservations.

Comet ML provides an end-to-end AI developer platform with two core product lines: Opik, an open-source GenAI observability and evaluation platform for tracing LLM calls, running automated evaluations, and optimizing agents; and a MLOps platform for experiment tracking, model versioning, dataset management, and production monitoring of traditional ML models.

Anyscale Platform is a fully managed, production-grade AI compute platform built on Ray—the open-source distributed runtime co-created by Anyscale's founders at UC Berkeley. It provides a unified environment for the complete AI/ML development lifecycle: large-scale multimodal data curation, distributed model training across thousands of GPUs, batch embedding generation, post-training (including RL and RLHF), and online inference serving. The platform exposes Python APIs that let developers scale existing code from a laptop to a multi-node cluster without rewrites, and supports flexible deployment as a hosted service or inside a customer's own VPC (BYOC) on major clouds and Kubernetes environments.

Helicone is an open-source AI gateway and LLM observability platform that lets developers integrate in minutes by pointing their existing OpenAI SDK to Helicone's proxy URL. It combines a unified multi-provider gateway (100+ models, automatic fallbacks, semantic caching, rate limiting) with full-stack observability (request logs, cost tracking, latency metrics, agent session tracing, prompt versioning, and user analytics) in a single platform, deployable as SaaS or self-hosted.

LiteLLM is an open-source AI gateway (proxy server) and Python SDK that gives developers and platform teams a single, OpenAI-compatible endpoint to access and govern 100+ LLM providers. Core capabilities include multi-provider routing with fallbacks, virtual key management, fine-grained cost and spend tracking per key/user/team/org, rate and budget enforcement, LLM guardrails, and integrations with observability tools. It also functions as an MCP gateway and A2A agent gateway. The enterprise edition adds SSO, audit logs, custom SLAs, and professional support.

Fireworks AI is an AI inference cloud and model lifecycle platform that lets engineering teams run, fine-tune, and scale open-source generative AI models in production. Built by the creators of PyTorch, it offers a serverless API across 100+ models, dedicated GPU deployments, and advanced tuning capabilities—including supervised, reinforcement, and quantization-aware fine-tuning—all behind an OpenAI-compatible interface with enterprise-grade security and global infrastructure.

Replicate is a serverless AI model hosting and inference platform that lets developers run, fine-tune, and deploy open-source and proprietary machine learning models with minimal code. Its core value proposition is eliminating GPU infrastructure complexity—developers call a unified API to execute models on managed cloud hardware that auto-scales to zero when idle. The platform's model marketplace hosts 50,000+ models from community contributors, AI labs (Anthropic, OpenAI, Google, ByteDance), and open-source projects; Cog standardizes custom model packaging into reproducible containers; and dedicated Deployment endpoints serve production workloads requiring guaranteed performance and isolation.

Together AI provides a full-stack 'AI Native Cloud' purpose-built for AI model deployment and development, combining serverless and dedicated LLM inference across 200+ open-source models, NVIDIA GPU cluster compute from H100 through Blackwell GB300, fine-tuning, evaluations, a code sandbox, and managed storage — all underpinned by proprietary systems research in inference optimization (FlashAttention, ATLAS, ThunderKittens) and an OpenAI-compatible API surface.