Services
Curated infrastructure, data, security, and API services that power production AI agents. Not a marketplace catalog — a shortlist of tools we would actually wire into a stack.
Showing 69 of 69 services
Modal: Serverless GPU Compute
Run training, fine-tuning, and inference workloads on serverless GPUs without managing infrastructure.
Robusta: AI Agent Security Audit
External security review for autonomous agent tool permissions, prompt injection risks, and deployment guardrails.
Pinecone: Managed Vector Database
Managed vector database for RAG, agent memory, and semantic search at scale.
Firecrawl: Web-to-Markdown for Agents
Turn any website into clean, structured Markdown or structured data for agent pipelines.
Scale AI: Data Labeling and RLHF
Human-in-the-loop data labeling and RLHF for training and fine-tuning custom models.
RunPod: GPU Cloud
Rent GPU compute by the hour or second for inference, training, and fine-tuning workloads.
ElevenLabs: AI Voice
Text-to-speech, voice cloning, and real-time voice agents for interactive AI experiences.
LiteLLM: Universal LLM Gateway
One API for 100+ LLM providers with load balancing, fallbacks, rate-limiting, and spend tracking.
Browserless: Headless Browser as a Service
Managed headless browsers for web scraping, form-filling, screenshots, and agent-driven browser tasks.
Composio: Tooling Layer for Agents
Pre-built, authenticated tool integrations so agents can act across SaaS apps, APIs, and databases.
Tavily: Search API for AI Agents
Real-time web search optimized for LLMs — returns clean, sourced answers instead of raw SERPs.
OpenRouter: Unified API for 100+ LLMs
One API key, one endpoint, 100+ models. OpenRouter handles routing, fallbacks, and model availability so your agent never stops because a provider is down.
Unstructured.io: Document Parsing for RAG
Extract clean, structured text from PDFs, Word docs, images, and HTML — ready for embedding and retrieval.
Portkey: AI Gateway + Guardrails
Route, observe, and secure LLM requests with enterprise-grade guardrails and model management.
Unkey: API Key Management for Agents
Managed API key infrastructure with rate limits, RBAC, usage analytics, and automatic key rotation.
Chroma: Developer-First Vector Database
Open-source vector database designed for AI applications with embeddings, full-text search, and metadata filtering.
Mem0: Memory Layer for AI Agents
Drop-in memory infrastructure that lets agents remember users, facts, and context across sessions.
Weights & Biases Weave: Agent Evaluation & Tracing
Trace, evaluate, and iterate on LLM agents with experiment tracking, prompt versioning, and production monitoring.
n8n: Workflow Automation with Agent Nodes
Open-source workflow automation with native AI nodes for building agent-driven processes without heavy code.
Replicate: Cloud API for Open-Source Models
Run open-source image, video, audio, and language models from a single API without managing GPUs.
Supabase: Postgres + Vector Extension for Agents
Open-source Firebase alternative with managed Postgres, pgvector, auth, and realtime — a common backend for agent apps.
Browserbase: Headless Browser Infrastructure
Managed headless browser platform for agent-driven web tasks — scraping, screenshots, form filling, and session persistence at scale.
Inngest: Durable Workflows for Agents
Reliable background jobs and durable execution for long-running agent tasks with retries, scheduling, and event-driven orchestration.
LanceDB: Serverless Vector Database
Open-source, serverless vector database with columnar storage and native Python/JS SDKs. Good for RAG and agent memory with no separate server to manage.
Langfuse: Open-Source LLM Observability
Trace LLM and agent runs, score outputs, manage prompts, and debug multi-step tool loops — self-host or cloud.
Modal: Serverless GPUs for Inference and Agents
Run GPU inference, fine-tunes, and agent workers as serverless Python functions — scale to zero, pay for compute seconds.
Daytona: Secure Sandbox Environments for AI Agents
Programmable sandboxed environments where agents can execute code, run tools, and operate filesystems in isolation — without risking your production infrastructure.
Helicone: LLM Gateway, Logging, and Cost Controls
Open-source-friendly gateway for routing, logging, caching, and cost controls across OpenAI-compatible providers.
Blume: Zero-Config AI-Ready Documentation
An open-source, MIT-licensed documentation framework that generates AI-ready docs from a Markdown folder — no config, no build step, just write and ship.
Cursor Router: Request-Level Model Routing
Cursor's classifier that routes each coding request to the optimal model based on query complexity — cutting AI coding spend 30-60% by matching routine work to cheaper models.
smolagents: HuggingFace's Code-First Agent Framework
A deliberately tiny agent framework where agents write Python code as actions instead of emitting JSON tool calls. ~1,000 lines of core logic, model-agnostic, Hub-integrated.
Fiddler AI: Observability and Security for LLMs
Enterprise-grade AI observability platform with real-time guardrails, trust scoring, and custom evaluators for agentic and LLM systems. Unified monitoring for predictive ML and generative AI.
Mastra: TypeScript-First AI Agent Framework
Production-grade TypeScript framework for building AI agents with workflows, memory, evals, and observability. Apache 2.0 open source with managed platform tier.
Microsoft Agent Framework 1.0
The production-ready merger of AutoGen and Semantic Kernel into a single SDK for building multi-agent workflows in .NET and Python.
Fly.io Sprites: the Ollama key stays on the org
A Sprite is a persistent Linux VM. The October 13 Ollama post says the gateway adds your key. The Sprite only sees a URL and a connection id.
IBM Bob self-hosted
IBM's announcement says the self-hosted deployment of Bob is generally available. The fetched page does not list a price or an install command.
OpenAI Decisions API
Public beta as of October 6, 2026. POST /v1/decisions on gpt-6-luna returns a probability, a choice, or a score. The decisions guide prices that endpoint at $0.10 per million input tokens and no output charge.
Gimlet Labs: Multi-Silicon Inference Cloud for Agentic AI
A multi-silicon inference platform that disaggregates AI models to run each phase of inference on the most appropriate chip — GPUs, CPUs, near-memory compute, and dataflow accelerators — built for low-latency agentic workloads.
OpenAI Agents API: Managed Codex Harness as a Product
OpenAI's public-beta API that exposes the Codex harness — session orchestration, context compaction, subagents, and sandboxes — as a fully managed service behind a single API call.
Brig: microVM Sandbox for AI Coding Agents
Apache 2.0 open-source microVM that runs AI coding agents inside a secure, low-overhead isolation layer on Mac or Linux — hardware-enforced boundaries for Claude Code, Codex, Cursor, and more.
WSO2 Agent Manager: Sovereign Agent Control Plane
Open-source control plane that governs AI agents across any framework, model, or deployment — sandboxed runtime, verifiable agent identity, and MCP-level governance.
AIR Security: Inline Firewall for AI Agents
Sequoia/Greenoaks-backed inline firewall that discovers and screens every skill, plugin, MCP server, and add-on entering an agent's context before the agent acts — blocking malicious instructions, untrusted data, and compromised tools at runtime.
Sonar Vortex: Semantic Code Navigation for Coding Agents
SonarSource's enterprise harness that replaces grep-and-read agent navigation with a live Unified Dependency Graph (SemSitter), cutting coding-agent token cost up to 36% and catching structural call sites that text search misses.
Diagrid Catalyst 2.0: Durable and Verifiable Agent Execution
Add automatic failure recovery and cryptographic execution attestation to AI agents built on LangGraph, Google ADK, Microsoft Agent Framework, and six other frameworks — without rewriting code.
Kong AI Gateway 2.0: Governed MCP, A2A, and LLM Traffic
Kong's GA AI gateway with MCP Server Bundling, identity-aware policies, and native support for Microsoft Foundry, SageMaker, and Bedrock AgentCore.
DigitalOcean M.A.R.S.: Managed Agents Runtime Services
A fully managed runtime for coding agents and long-running agentic workflows — durable sessions in Firecracker microVMs, governed tool access via Action Gateway, and harness flexibility for Claude Code, Codex, LangGraph, and CrewAI.
Temporal Agent Harness: Durable Agent Infrastructure
An open-source outer harness that wraps your existing agent SDK with durable execution, approval policies, typed operations, and structured event streams. Every agent becomes a Temporal Workflow that survives crashes, deployments, and multi-day waits.
Cloudflare Kitesurf: Agent-First Browser
Cloud-hosted browser built specifically for AI agents — runs in V8 isolates instead of Chromium, using 3-7x less CPU and memory per screenshot. Free during beta.
Cloudflare Wallets: Programmable Agent Payments
Stablecoin wallet system for AI agents — Account Wallets for humans, Virtual Wallets for agents with spend caps. Built on the x402 protocol with cloudflare.pay handles.
agentgateway: Kubernetes-Native AI Gateway for MCP and A2A Traffic
A Rust-based, Linux Foundation-governed gateway that unifies MCP tool routing, A2A agent communication, LLM inference proxying, and Kubernetes Gateway API — purpose-built for production agentic systems.
Northflank: Full-Stack AI Sandbox and Agent Infrastructure Platform
MicroVM-backed sandboxes, databases, APIs, CI/CD, and GPU workloads in one platform — deploy in your own cloud (BYOC) or Northflank's managed cloud.
osModa: Self-Healing AI Agent Hosting
AI-native operating system built on NixOS and Rust for autonomous agent hosting — watchdog auto-restart, atomic rollbacks, and tamper-proof audit logging on dedicated servers.
vLLM Semantic Router: Mixture-of-Models Routing
Open-source intelligent router that classifies LLM requests and routes them to the right model — semantic caching, safety filtering, and cost optimization across heterogeneous backends.
Astrix Security: NHI Governance & AI Agent Control Plane
Discover, secure, and deploy AI agents and non-human identities at scale — the leading platform for agent identity governance, access control, and threat detection.
Crossmint: Agentic Payments Infrastructure
Give your AI agents wallets, stablecoin rails, and card payment capabilities — programmable spending limits, multi-chain support, and Visa/Mastercard integration in one API.
Ferrogate: Self-Hosted AI Gateway
Open-source Rust/Pingora AI gateway for self-hosted LLM traffic control — provider routing, virtual API keys, budgets, caching, MCP tool execution, and observability.
MCP Memory (libSQL): Persistent Agent Memory
High-performance persistent memory system for Model Context Protocol powered by libSQL — vector search, semantic knowledge storage, and relationship management for AI agents.
AWS Strands Agents
An open-source, model-driven agent SDK from AWS — define a model, tools, and prompt; the LLM handles planning and execution. Apache 2.0, Python and TypeScript.
Code Atlas
A code intelligence graph that gives AI coding agents deep, token-efficient understanding of your codebase — structure, docs, and dependencies in one searchable graph. MCP-compatible, self-hosted, Apache 2.0.
DeepInfra: Low-Cost Inference API
Usage-based inference cloud with OpenAI-compatible API, some of the cheapest per-token pricing in the market, and on-demand GPU rental.
Baseten: Production Inference for Open Models
Serverless per-token Model APIs plus dedicated GPU-minute deployments for open-weight models. Now a HuggingFace inference provider. Strong for custom model serving, expensive for always-on workloads.
Muse for Small Business
Meta's September 29, 2026 expansion of Muse: business connectors, a publish/send/spend approval gate, and no price on the launch post.
Pencil Scribble
Pencil's public launch of Scribble on September 28, 2026: a brief goes in, the marketer reviews the plan before it runs, and media buying is not live yet.
Thrad Autopilot
Ads inside AI conversations, with a submit gate and a real price — and two pages that do not agree on who can use it.
Multiply Ad Spend Recovery Agent
A September 28, 2026 Google Ads agent that proposes negatives, pauses, and bid changes, and will not apply a spend-affecting action without approval.
Google Flow Music Spaces
Google's October 6, 2026 note: natural-language Spaces can export as VST3/AU plugins. The post does not list a price or an export click.
LightOnOCR-3
Three Apache-2.0 OCR models. An empty prompt transcribes. The word grounding adds boxes, image descriptions, and chart tables. No price on the pages fetched.
ChatGPT Slides: ask in Work, inside Space
The October 16 release notes say Slides is rolling out to eligible Pro, Business, and Enterprise users with Space access. Start by asking in Work. The notes do not print a separate price.
Notion seat credits: 1,500 Standard, 500 Select
The October 19 blog puts 1,500 Standard credits and 500 Select credits on Business and Enterprise seats. New customers were on the system that day. Existing customers move on November 2. The connect click is not named.