Agent Frameworks Landscape 2026
The 13 published agent frameworks filed under ai agent frameworks, positioned on the same axes as the whole category: public evidence and search-interest trend, measured against every ai agent frameworks product rather than only against each other.
Scroll the chart sideways to read every position, or browse the complete list below. Each tool links to its review, pricing, alternatives, and comparisons.
Where Each Tool Lands
Strong & Gaining (3)
- LangChainFree tier
Agent Framework
Public signal strength: 100th percentile · Interest trend: 71st percentile in category
GitHub stars:148.5k - LangGraphFree (open source)
Agent Framework
Public signal strength: 71st percentile · Interest trend: 57th percentile in category
GitHub stars:43.9k - CrewAIFree tier
Agent Framework
Public signal strength: 57th percentile · Interest trend: 100th percentile in category
GitHub stars:60.1k
Quieter & Easing (4)
- LlamaIndexFree tier
Agent Framework
Public signal strength: 43rd percentile · Interest trend: 14th percentile in category
GitHub stars:52.7k - OpenAI Agents SDKFree (open source)
Agent Framework
Public signal strength: 29th percentile · Interest trend: 0th percentile in category
GitHub stars:30.3k - Google Agent Development KitFree (open source)
Agent Framework
Public signal strength: 14th percentile · Interest trend: 43rd percentile in category
GitHub stars:22.0k - PydanticAIFree (open source)
Agent Framework
Public signal strength: 0th percentile · Interest trend: 29th percentile in category
GitHub stars:20.9k
Not yet plotted (6)
Listed alphabetically. A product appears here when its public evidence does not meet the positioning threshold, when its search interest is too small to read a direction from, when it falls outside the category chart limit, or when the product is discontinued. Missing evidence is shown as missing, never converted to zero. A tool we could not measure is listed without a position rather than placed at the bottom of the scale.
How to read this view
A tool must show measured activity on at least 2 different platforms, at least one of which must be a primary source (Google Trends, GitHub, Docker Hub, npm, PyPI, Hugging Face and Stack Overflow); Hacker News and Product Hunt can supply the second. The horizontal position shows that evidence: Measured activity on each qualifying platform (Google Trends, GitHub, Docker Hub, npm, PyPI, Hugging Face, Stack Overflow, Hacker News and Product Hunt), log-normalized and percentile-ranked within the category. Each platform counts once and is capped, so breadth of evidence counts for more than a single large number. The vertical position shows A tool's share of its category's Google Trends search interest, and which way that share moved over the last three months. Gaining share does not mean searches grew; it means this tool held more of the category's attention than before. Interest too small to read reliably is reported as unmeasurable, so those tools are listed rather than positioned. Both percentiles are computed across the whole ai agent frameworks category, so a product keeps the same position here as on the category chart. This measures how much verifiable public evidence exists for a tool. It is not a measure of product quality, market share, customer count, or enterprise adoption. Missing evidence is shown as missing, never converted to zero. A tool we could not measure is listed without a position rather than placed at the bottom of the scale.