Today's take
The strongest signal is growing attention to the infrastructure around agents: execution environments, web access and engineering practices. Niko1221/Strata and lexmount/moli hold the top two open-source positions, while affaan-m/ECC climbs from #21 to #6. The emphasis is moving toward making agents easier to deploy and more dependable at completing work.
Coding tools also appear among both app usage leaders and rising apps, providing a directional connection to the open-source tooling trend. The snapshots cover different dates: October 4 for open source and October 2 for models and apps. Usage figures cover only traffic through OpenRouter with public attribution; Tokens measure usage, not people.
Open source
Three themes stand out across the Top 50: local execution, web data access and reusable workflows. Together, they address how agents run, acquire information and deliver results. Stars and ranking changes indicate attention, rather than demonstrated production adoption.
- Local execution and web interaction: #1 Niko1221/Strata adds 2,026 stars in 1 day, reaching 7,367, with local inference and compatible APIs for consumer hardware. #2 lexmount/moli adds 1,783 stars and moves from #5 to #2 with an agent-focused headless browser. They address separate barriers to running agents and operating on the web.
- Web access as a foundational capability: #3 Panniantong/Agent-Reach adds 1,325 stars with a unified CLI for reading and searching across platforms. D4Vinci/Scrapling rises from #49 to #30 with 279 additional stars, reinforcing interest in retrieval and crawling as dedicated tools.
- Engineering practices become reusable assets: #6 affaan-m/ECC adds 788 stars and climbs 15 places, combining skills, memory, security and development practices in an agent engineering system. #7 mattpocock/skills adds 677 stars, supporting the broader shift toward modular engineering guidance.
- #22 heygen-com/hyperframes: The day's new Top 50 entrant adds 382 stars, reaching 56,139. Its HTML-to-video approach explicitly targets agents, packaging media delivery as a callable capability and extending workflows beyond code.
Models
Growth among usage leaders differs from acceleration outside the top ten. Several leading models carry a batch label, but the rankings do not identify the apps or tasks responsible for their Tokens.
- Usage leaders: Space Bunny Alpha leads with 5.89T Tokens, up 14% versus the previous day. DeepSeek V4.1 Flash (batch) records 3.48T, up 3%, and GLM 5.3 Flash (batch) records 1.61T, up 12%.
- Fastest movers: Among the Top 50 outside the top ten, Hy4 preview records 61.7B Tokens and +219%; GPT-6 Luna Pro (batch) records 214B and +117%; Claude Opus 5 (batch) records 121B and +94%.
- Divergence within a model family: GPT-6 Luna (batch) records 729B Tokens and -27%, contrasting with growth in GPT-6 Luna Pro (batch). This establishes different usage trajectories, but does not demonstrate traffic migration.
Apps
Coding agents are central to this publicly attributed usage ranking. The top five include three CLI agents, one IDE extension and one personal agent, while the rising list also favors coding tools. Growth percentages are not supplied, so the magnitude of acceleration cannot be quantified.
- Highest usage: Hermes Agent leads with 2.01T Tokens, followed by Claude Code at 1.23T, Kilo Code at 1.14T, Cline at 1.06T and Codex at 1.01T.
- Growth direction: Relative to the average usage of the preceding three equal-length periods, Freebuff ranks first on the rising list with 656B Tokens in the current period. Codex, omp, Claude Code and OpenHands follow; the latter two record 1.23T and 130B respectively.
What to watch
Open-source attention and app usage both point toward more engineered agent workflows. Watch whether tooling and skill projects sustain their momentum and coding apps continue growing. Establishing model migration will require app-to-model attribution data.
