Skip to content
Agent Search Engine.
Submit a tool
Menu

Report · Issue 01 · Catalogue study

The State of AI Agents

What is in ASE's index: 593 records across agents, frameworks, connectors and related software. Explore the recorded categories and repository signals, with the definitions and limits beside the numbers.

By Agent Search Engine · Editorial revision 2026-10-08 · Latest recorded metrics 2026-10-26

593

Catalogue records

One primary category and entity type per record.

63%

Recorded open-source license

376 of 593 records name recognized software licenses; current product terms still need checking.

357

Declared code language

Among records with recognized open-source software licenses.

414

Recorded push date

An activity field, not a release or maintenance guarantee.

Figure 1

Recorded software-license evidence

Recognized software licenses are counted separately from prices. Missing, documentation-only and unrecognized license values remain unchecked; a public repository does not establish a product's software license.

  • Recognized open-source license376 records376 ÷ 593 × 100 ≈ 63%
  • Source-available license5 records5 ÷ 593 × 100 ≈ 1%
  • Mixed licenses1 records1 ÷ 593 × 100 ≈ 0%
  • Software license not checked211 records211 ÷ 593 × 100 ≈ 36%

n = 593 records

Figure 2

Recorded pricing classifications

These labels describe the index's current records. They do not establish total operating cost or mutually exclusive business models in the market.

  • Costs not checked (legacy category)408 records408 ÷ 593 × 100 ≈ 69%
  • Freemium74 records74 ÷ 593 × 100 ≈ 12%
  • Paid product83 records83 ÷ 593 × 100 ≈ 14%
  • Free offering19 records19 ÷ 593 × 100 ≈ 3%
  • Unknown pricing9 records9 ÷ 593 × 100 ≈ 2%

n = 593 records

Figure 3

Primary repository language

Language counts describe implementation, not supported spoken languages, available SDKs or what a user needs to know.

  • Python158 records158 ÷ 357 × 100 ≈ 44%
  • TypeScript112 records112 ÷ 357 × 100 ≈ 31%
  • Go26 records26 ÷ 357 × 100 ≈ 7%
  • JavaScript21 records21 ÷ 357 × 100 ≈ 6%
  • Rust12 records12 ÷ 357 × 100 ≈ 3%
  • Swift6 records6 ÷ 357 × 100 ≈ 2%
  • Other (9 languages)22 records22 ÷ 357 × 100 ≈ 6%

n = 357 records with recognized open-source software licenses and a declared primary language

Figure 4

Records by primary category

One primary category per record keeps this count additive. It does not capture every capability a product offers, nor establish that two products are substitutes.

n = 593 records

Figure 5

Change in recorded GitHub stars

Only projects observed in both snapshots are compared. A newly listed repository has no measured earlier baseline. Stars measure interest, not product usage, revenue or quality.

OpenCode222k stars▲ 4,560ruflo79k stars▲ 2,560firecrawl195k stars▲ 2,551Hermes Agent257k stars▲ 2,117Apify MCP Server13k stars▲ 1,261pentagi32k stars▲ 1,223Scrapling90k stars▲ 1,206gemini-cli108k stars▲ 950Windows-MCP10k stars▲ 862cua31k stars▲ 802

19 October – 26 October 2026 · 7 days between recorded snapshots

Reproduce a fixed sample.

The downloadable sample contains 454 public records with metrics as of 5 October 2026. It preserves the original recorded fields and grouped counts. The charts above use the cached catalogue at page generation and the current software-license check, so they may differ after later refreshes or classification corrections. Do not combine the fixed export with a different chart date as though they were one dataset.

Download 454 records · CSVDownload data and manifest · JSON
  1. Count rows excluding the CSV header, or count the JSON records array.
  2. Group by the exact pricing, category or entity_type value. Keep missing values separate. The legacy pricing value open-source does not establish a software license or operating cost.
  3. For a share, divide the group count by the stated denominator and multiply by 100. Displayed whole percentages are rounded and may sum to 99% or 101%.
  4. The live code-language chart now includes records with recognized open-source software licenses and a recorded language. The export's original pricing-based grouping does not reproduce this corrected chart.
  5. The JSON manifest includes the source snapshot's SHA-256 and original grouped counts. These make the fixed sample checkable without an account; they are not a current license audit.

What these data cannot tell you

  • Which tool is most accurate, easiest, safest or best for your task. That needs a task-specific evaluation.
  • How many people use a product. GitHub stars are repository signals, and several tools can share a repository.
  • Whether an active push fixes bugs or whether an inactive repository is unsuitable. Activity alone does not establish maintenance quality.
  • Where every model, file or log runs. Self-hosting one component does not prove an entirely local system.
  • Global consumer demand or an entire market's growth. Changes in ASE's coverage can change the sample.

The strongest limitation is selection bias: a collection built around repositories will naturally contain many repositories. An independent, representative sampling method could produce a very different distribution.

Methods, updates and citation

Catalogue fields are collected from public repositories and reviewed product sources. Scheduled snapshots update selected metrics; page caching can delay display. Upstream repository pushes, record verification and article revision dates describe different events. Rankings and report inclusion are independent of sponsorship.

Cite “Agent Search Engine, The State of AI Agents, Issue 01,” with the metric date shown above or the fixed export date, and the retrieval date. Our report and export are available under CC BY 4.0; that does not relicense vendors' software or documentation.

Full methodologyEvaluate a tool yourselfReport a data correction