Bookmarks

AI Stack Radar — November 11, 2026

Agents & Orchestration

Dify

Builds agents and workflows on a canvas, then publishes the result as a web app, an API, an embed or an MCP-compatible tool. The knowledge pipeline (extract, chunk, index) is part of the same editor rather than a separate product. **Replaces:** a hand-rolled LangChain service plus its admin UI. **Pricing:** free self-hosted Community Edition; Cloud plans scale with team size and usage.

n8n

Visual workflow runner with 500+ native integrations, any HTTP API, and human approval steps you can drop into an agent loop. Every run keeps a log you can open and replay. **Replaces:** Zapier for anything with branching, or a pile of cron scripts. **Pricing:** self-host for free, or pay for the hosted cloud version.

Lindy

Hands you an "employee" wired into Slack, Gmail and Calendar that chases overdue invoices, answers new leads and builds client reports. It's closer to a ready-made assistant than to a builder. **Replaces:** a virtual-assistant retainer for inbox chores. **Pricing:** free tier with $50 in credits that expire after 7 days; paid plans start at $29.99 per user per month.

Mem0

Two calls, `add` and `search`, store facts from conversations and pull them back later, so an agent remembers that a user is vegetarian without replaying the whole chat. **Replaces:** a home-grown vector table of "user preferences". **Pricing:** hosted API with a free start; limits are on its pricing page.

MCP & Tool Plumbing

Zapier MCP

Points Claude, ChatGPT or Cursor at Zapier's 9,000+ apps, with Zapier holding the credentials and access rules. Each tool call costs two tasks from your existing quota. **Replaces:** writing and hosting one MCP server per SaaS. **Pricing:** included in every Zapier plan, billed against tasks.

PulseMCP

A catalog of MCP servers, clients and real use cases, with a weekly newsletter. New submissions are paused right now, so treat it as a browsing tool rather than a place to list your own. **Replaces:** scrolling GitHub topic pages to find a server. **Pricing:** free to browse.

Coding & Dev

Cline

Coding agent that makes coordinated edits across files and runs terminal commands, on Claude, GPT, Gemini or a local Ollama model. It ships as an editor extension, a desktop app, a CLI and an SDK, so the same agent can run in CI. **Replaces:** a locked-in IDE assistant. **Pricing:** open source; the plan page covers the hosted options.

Zed

A fast native editor where the AI is optional. Pro is $10 a month and includes $5 of hosted tokens plus unlimited edit predictions, then bills overage at API price plus 10%. **Replaces:** VS Code with a Copilot subscription. **Pricing:** free personal plan (bring your own keys); Pro $10/month with a 14-day trial.

Model Hosting & Inference

Together AI

Runs open-source models behind serverless, batch, provisioned and dedicated endpoints, and also rents GPU clusters and code sandboxes. The batch tier suits overnight jobs that don't need an answer now. **Replaces:** renting your own GPUs to serve Llama-class models. **Pricing:** pay per token; no long-term commitment on serverless.

Fireworks AI

One endpoint for open-weight models across 30+ regions, with routing that picks a model by accuracy against cost, and training on your own data. Some models list at $0.15 per million input tokens. **Replaces:** juggling several inference vendors. **Pricing:** per-token, listed model by model.

Replicate

A cloud API over thousands of community-published image, video, speech and language models. You can also package your own with Cog and deploy it, and nothing is billed while traffic is zero. **Replaces:** a GPU box you forgot to turn off. **Pricing:** billed per second of compute; sign-up is free.

Modal

Python-first cloud for inference endpoints, fine-tuning, and isolated sandboxes for agents, with GPUs that scale up and back down on their own. **Replaces:** Kubernetes plus a GPU node pool. **Pricing:** per-second billing with $30 of free compute each month.

Image & Design

Krea

Generates, edits and upscales images, video and 3D in the browser, with real-time rendering as you draw and LoRA training for your own style. **Replaces:** juggling a generator, an upscaler and a style-tuning tool. **Pricing:** free plan with 100 daily units; paid from $9/month.

Photoroom API

Product-photo edits over plain HTTP: background removal, AI shadows, padding, lighting fixes and upscaling. A catalog pipeline is a loop over this endpoint. **Replaces:** a freelancer cutting out SKUs by hand. **Pricing:** 1,000 free sandbox calls a month (watermarked); paid plans bill per image.

fal.ai

One key and one SDK cover image, video, audio and 3D models, and the same account hosts your own models on serverless GPUs. **Replaces:** separate accounts at five generation vendors. **Pricing:** per-output prices from prepaid credits.

Video & Motion

Luma

Agents that research, generate and revise video, images, audio and text while holding one brand's style steady across all of them. Useful when the pain is consistency, not a single pretty clip. **Replaces:** a stock-footage subscription plus an editor pass. **Pricing:** free trial; tiers sit behind sign-up.

HeyGen

Turns a script, image, audio file or slide deck into a talking-avatar video, including an avatar of you, and translates it into 175+ languages with lip-sync. **Replaces:** filming the same training video eight times. **Pricing:** free plan for text-to-video; paid plans unlock editing and exports.

Voice & Audio

AssemblyAI

Speech-to-text for recorded and live audio, with add-ons for PII redaction and structured extraction, plus a Voice Agent API that does speech-to-speech over a single WebSocket. **Replaces:** stitching an STT, an LLM and a TTS together yourself. **Pricing:** usage-based; details on its pricing page.

Vapi

Assembles a phone-capable voice agent: you pick the voice and the conversation flow, and it handles telephony with Twilio and others and watches calls live. **Replaces:** a call-center script and an IVR tree. **Pricing:** usage-based; see its pricing page.

LiveKit

WebRTC edge network plus an agent framework for voice, video and robotics agents, with phone numbers, SIP and hosted STT, LLM and TTS models. **Replaces:** running your own media servers. **Pricing:** free account with 1,000 agent session minutes a month, no card.

Research & Retrieval

Firecrawl

Searches the web, scrapes pages into Markdown or JSON, crawls whole sites, parses PDFs and watches pages for changes. **Replaces:** a Playwright scraper you keep patching. **Pricing:** free plan with 1,000 credits a month; Hobby is $19/month.

Tavily

Search API built for agents: it returns extracted, model-ready content, not a list of blue links, and can also crawl and map a site. **Replaces:** Bing or Google search scraping inside an agent tool. **Pricing:** free to start; limits and paid rates on its pricing page.

turbopuffer

Vector and full-text search database that keeps data on S3 behind an SSD cache, so idle collections cost almost nothing. A 1M-document sample works out to about $7 a month. **Replaces:** an always-on Pinecone or pgvector instance. **Pricing:** usage-based, with a $16/month minimum on the launch plan.

Zep

Syncs Slack, Drive and Snowflake into a governed context graph that agents query at runtime, with attribute-based access rules and an audit log of every request. **Replaces:** a shared RAG index with no permissions. **Pricing:** free tier listed; enterprise runs in managed cloud, your keys or your VPC.

Local & Self-Hosted Inference

Ollama

Runs open models on your own machine, or on its cloud when the model won't fit, and says it doesn't store or train on prompts. **Replaces:** hand-built llama.cpp setups for day-to-day use. **Pricing:** free local use; cloud models are on its pricing page.

LM Studio

Desktop app whose new Bionic agent writes documents, edits code and drives the computer, on local models or hosted open ones, with on-device voice transcription. **Replaces:** a chat UI bolted onto a model server. **Pricing:** free app download; an account is optional.

Evals & Observability

Langfuse

Traces LLM calls, tool use and retrieval steps, adds evals and prompt management, and has an optional gateway that tracks cost per user. **Replaces:** printing prompts to a log file. **Pricing:** free Hobby cloud tier (50k observations a month); MIT-licensed if you self-host.

Braintrust

Logs production agent traces, scores them with LLM, code or human graders, and surfaces recurring failure patterns. Real traces become eval datasets in a click. **Replaces:** a spreadsheet of "bad answers" someone keeps by hand. **Pricing:** self-serve sign-up; tiers on its pricing page.

Promptfoo

Red-teams AI apps and agents for prompt injection, jailbreaks, PII leaks and unsafe tool use, then reports in CI and on pull requests with suggested fixes. **Replaces:** a one-off pentest before launch. **Pricing:** open-source edition to start; enterprise plans separate. --- *29 products this week. Found a dead link or have something to add? Reply to the email.*