AI Agent Infrastructure
The model is the easy part. Shipping an agent means routing calls across providers, seeing what it did, giving it memory, feeding it clean data and giving it somewhere to run. These are the pieces, grouped by job, each with one line on what it does. 27 tools listed.
See also: AI gateways best-of list · MCP Servers & Tools
Models, Inference and Gateways
One API in front of many models, with spend limits, failover and caching, or inference you run yourself.
Hopscotch
One API for 500+ models with spend limits per key, teammate and workspace, plus fallbacks and request logs.
Opper
EU-hosted gateway to 700+ models behind an OpenAI-compatible API, with spend caps and zero data retention.
Bifrost
Open-source gateway with failover, load balancing, semantic caching, budgets and an MCP gateway built in.
Navige
Sits between your agents and their models and cuts spend with caching, smart routing and loop guards.
RunInfra
Open-model APIs built for agents, with warm replicas and one key that works with the OpenAI and Anthropic SDKs.
Inferock
Watch your own OpenAI, Anthropic or Gemini calls for failures and overbilling, or run inference with credits back on eligible failures.
Lumabri
Splits a model's layers across your own computers on a LAN, so models too big for one machine still run privately.
Trace, Evaluate and Track Spend
See what agents did in production, score them before launch, catch leaked keys and attribute every dollar.
Latitude
Open-source tracing for agents that turns production failures into tracked issues with evals attached.
Promptic
OpenTelemetry-native LLM tracing, model benchmarks and prompt and agent tuning scored on your own quality and cost metrics.
Cekura
Simulates conversations across personas before launch and monitors live voice and chat agents for tool-call errors.
Weir
Plants fake keys, runs your agent's flows and reports whether they leaked into traces, logs or the model.
Costura
Self-hosted spend attribution across OpenAI, Anthropic, OpenRouter, Cursor and more, down to the key and the person.
Memory and Knowledge
Places for agents to keep context, share records with people and read facts they can trust.
Maximem Synap
Memory layer with entity resolution and temporal reasoning, plus native support for LangChain and the Claude Agent SDK.
Busabase
Open-source system of record where agents and people share data and docs, every change keeps its history and key writes can wait for review.
NOAN
Your verified business facts behind one API over REST and MCP, read live by every agent, app and site.
Glider
Open-source vector database that writes durably to S3-compatible storage and ships an MCP memory server.
KiviDB
Multi-threaded Redis and Valkey alternative in Rust with full-text and vector search built in. Existing Redis clients work unchanged.
Documents and Web Data
Turn PDFs, scans, contracts and live web pages into clean Markdown or JSON an agent can use.
Context.dev
Scrape any URL into LLM-ready Markdown, search the live web and crawl sites, with SDKs, a CLI and an MCP server.
Invofox
Parses invoices, payslips, contracts and other documents into validated JSON or Markdown for RAG and LLM pipelines.
Sygal
Local-first app and CLI that turns PDFs, Office files, scans and web pages into clean Markdown or JSON.
Nace
A Perception API that reads any file and a Decision API for small, fast calls, under one key.
Runtimes, Browsers and Desktops
Somewhere for agents to run: managed agent APIs, stateful actors, and browsers and desktops they can drive.
Omnara
Open-source, self-hostable take on managed agents: run agents on any model and machine with tools, secrets and permissions built in.
ZooWork
Domain experts build in a Builder UI while developers ship through a Managed Agent API, with approvals and controls.
Durable Actors
Open-source alternative to Cloudflare Durable Objects: stateful actors, each with its own SQLite, for chat and multi-agent apps.
ego lite
Free macOS browser your agents drive in their own Spaces, using your Chrome logins while you keep browsing.
pw-repl
Playwright REPL you and your agents share, in your own logged-in browser or a headless one.
Braus
Gives agents their own Mac desktop to work in apps that have no API, connected to Claude Code or Codex over MCP.
Building infrastructure for AI agents? Featured listings sit at the top of their category for 30 days.
Get your tool featuredWant developers asking ChatGPT or Perplexity to land on your product? See AI search visibility