Agents & automation, with a verdict on each.
Tools that let software take multi-step actions on its own — browse, call APIs, run code, finish a job without a human in the loop for every step.
Most teams need one of these, not four. The differences that matter are how they fail and how much of your budget they burn when they do.
Moving fastest this week: ponytail (+23k), firecrawl (+16k), hermes-agent (+16k).
Adopt now
6Mature enough to put in production this quarter.
Turns any website into clean markdown an LLM can read, including pages that need JavaScript to render.
Instead of: A homegrown Playwright scraper plus the week per quarter spent fixing it.
A Python scraping library that adapts when a site changes its markup, instead of breaking silently.
Instead of: The recurring afternoon spent repairing selectors after a site redesign.
A drag-and-drop workflow builder that wires your apps together and can call an LLM at any step.
Instead of: Zapier, Make, and the pile of one-off scripts nobody maintains.
A polished self-hosted chat interface over your own models or API keys, with document upload and user accounts built in.
Instead of: Per-seat subscriptions on a hosted chat product.
Google's terminal coding agent — reads your repo, edits files, runs commands, and connects to MCP tools.
Instead of: Part of a contract developer's hours on routine changes and test writing.
Per-second monitoring for servers and containers that installs in one command and needs almost no configuration.
Instead of: The entry tier of Datadog or New Relic.
Pilot it
16Worth a timeboxed spike before you bet on it.
AI-powered desktop assistant
Instead of: manual AI model switching
CLI tool for AI agents to search internet platforms
Instead of: Manual social media monitoring or paid API services
Pre-LLM token compressor for logs, JSON, and tool outputs, saving 20-95% tokens.
Instead of: Manual prompt trimming or paying for larger context windows.
AI code review and suggestion tool
Instead of: manual code review and suggestion
Local-first design tool with AI coding agent
Instead of: Manual design tools or Claude Design
Lets an LLM drive a real browser — click, type, scroll — so it can use websites that have no API.
Instead of: A contractor manually copying data between systems every week.
Open-source agent and automation framework
Instead of: Manual agent development or paid automation tools
A document-understanding engine that parses messy PDFs and scans properly before answering questions about them.
Instead of: Paying a vendor per page for document extraction.
A visual builder for LLM apps: prompt chains, retrieval over your documents, and agent steps without writing the plumbing.
Instead of: Three weeks of a contractor building a bespoke RAG backend.
The most widely used Python toolkit for chaining model calls, tools and retrieval into an application.
Instead of: Writing your own provider adapters, retry logic and tool-calling loop.
SDK for building AI agents in Python & TypeScript
Instead of: Manual agent deployment scripts
Python agent for automation and system control
Instead of: Manual scripting or paid automation tools
A self-hostable chat interface that puts several model providers and a library of configured agents behind one login.
Instead of: Per-seat ChatGPT Team or Claude Team subscriptions across a whole company.
Framework to define reusable AI agent skills in Python.
Instead of: hand‑coded prompt scripts or ad‑hoc bot logic.
Open-source SRE agent for incident management
Instead of: Manual incident response or Jira workflows
AI agent for terminal automation
Instead of: Manual scripting and workflows
Watch, or skip
15Real software; just not where a small team's next hundred hours should go.
AI agent performance optimization system
Instead of: manual performance tuning
AI agent framework for automation
Instead of: manual automation scripts
LLM-based chatbot agent in Python
Instead of: manual chat support or paid chatbot services
Plugin-based automation framework
Instead of: Manual scripting or paid automation tools
AI coding assistant
Instead of: manual coding tasks
AI agent for automation
Instead of: manual automation tasks
Turns codebases into queryable graphs using ASTs, no vector stores, built for AI coding assistants.
Instead of: Manual grep/search, paid tools like Sourcegraph or GitHub Code Search.
AI-powered news aggregation and monitoring dashboard
Instead of: manual news tracking and monitoring tools
Framework for defining and sharing AI agent skills across Claude Code, Codex, and other coding agents.
Instead of: Custom prompt libraries, ad-hoc agent instructions, copy-pasted context files.
Stores and recalls AI agent conversation history using Claude and ChromaDB, but requires heavy setup and maintenance.
Instead of: Manual context logging or paid tools like LangChain + vector DBs for agent memory.
AI agent evolution engine
Instead of: manual agent tuning
Reinforcement learning for LLM agents
Instead of: Custom RL implementations
AI agent platform with memory and learning
Instead of: manual automation scripts
Multi-agent programming for LLMs
Instead of: Custom LLM integration code
Small AI framework for building agents
Instead of: Manual agent workflows
Agents & automation, in short
- What are the best open source AI agent frameworks right now?
- We would put firecrawl, Scrapling, n8n into production this quarter. firecrawl — Turns any website into clean markdown an LLM can read, including pages that need JavaScript to render. The full list below ranks 37 by momentum, size and how recently they shipped.
- Which of these can I use in a commercial product?
- 28 of the 37 carry a permissive licence (MIT, Apache-2.0, BSD). 9 do not — firecrawl (AGPL-3.0), n8n (Custom / check the LICENSE file), open-webui (Custom / check the LICENSE file) and others. Copyleft and source-available licences carry obligations when you ship them inside something you sell. Read the LICENSE file; this is a reading of the label, not legal advice.
- How is this list ranked?
- By real week-over-week star movement first, with a size floor so an established tool is not buried by a two-week-old project, and a hard penalty for anything that has stopped shipping. Verdicts are an editorial call for a team of 2-20 with limited engineering hours — not a judgement on the software, and not the same call we would make for a lab.
Which of these matters to your business?
Tell us what you sell in one sentence and we will hand you three specific moves — priced per month, with the arithmetic shown. Free, no signup.
Give me three moves