Apex Browse: Deterministic Browser Workflow Execution with Bounded Large-Language-Model Repair

Apex Browse demonstrates that an LLM need not remain in the loop for every browser action after it has expressed a complete workflow. A strict DSL, session-local Playwright executor, accessibility-oriented resolver, explicit postconditions, compact evidence, and candidate-bound repair can shift routine interaction to deterministic code while preserving a controlled path back to model reasoning. In the reported 150-attempt controlled experiment, Apex Browse completed 49 of 50 agent tasks and substantially reduced wall time, tokens, MCP calls, and serialized result bytes relative to official Playwright MCP. The strongest empirical conclusion is limited but useful: for the tested semantic workflows under the pinned configuration, deterministic-first execution delivered materially lower interaction cost with one observed planner-level failure. The benchmark does not establish universal web-agent superiority, visual competence, or production-grade security. The broader design principle is to reserve probabilistic reasoning for uncertainty and use deterministic systems for execution that can be fully specified and verified. Browser agents need both capabilities. Apex Browse places the boundary at failed semantic resolution and makes that boundary explicit, inspectable, and measurable.

[1] S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y. Cao, “ReAct: Synergizing reasoning and acting in language models,” in Proc. International Conference on Learning Representations, 2023. [Online]. Available: https://arxiv.org/abs/2210.03629 [2] X. Deng, Y. Gu, B. Zheng, S. Chen, S. Stevens, B. Wang, H. Sun, and Y. Su, “Mind2Web: Towards a generalist agent for the web,” in Advances in Neural Information Processing Systems, vol. 36, 2023. [Online]. Available: https://proceedings.neurips.cc/paper_files/paper/2023/hash/5950bf290a1570ea401bf98882128160-Abstract-Datasets_and_Benchmarks.html [3] S. Zhou, F. F. Xu, H. Zhu, X. Zhou, R. Lo, A. Sridhar, X. Cheng, T. Ou, Y. Bisk, D. Fried, U. Alon, and G. Neubig, “WebArena: A realistic web environment for building autonomous agents,” in Proc. International Conference on Learning Representations, 2024. [Online]. Available: https://proceedings.iclr.cc/paper_files/paper/2024/hash/4410c0711e9154a7a2d26f9b3816d1ef-Abstract-Conference.html [4] H. He, W. Yao, K. Ma, W. Yu, Y. Dai, H. Zhang, Z. Lan, and D. Yu, “WebVoyager: Building an end-to-end web agent with large multimodal models,” in Proc. 62nd Annual Meeting of the Association for Computational Linguistics, pp. 6864–6890, 2024, doi: 10.18653/v1/2024.acl-long.371. [5] A. Drouin, M. Gasse, M. Caccia, I. H. Laradji, M. Del Verme, T. Marty, D. Vazquez, N. Chapados, and A. Lacoste, “WorkArena: How capable are web agents at solving common knowledge work tasks?” Transactions on Machine Learning Research, 2024. [Online]. Available: https://openreview.net/forum?id=BRfqYrikdo [6] T. Le Sellier de Chezelles et al., “The BrowserGym ecosystem for web agent research,” Transactions on Machine Learning Research, 2025. [Online]. Available: https://openreview.net/forum?id=5298fKGmv3 [7] World Wide Web Consortium, “Accessible Rich Internet Applications (WAI-ARIA) 1.2,” W3C Recommendation, Jun. 2023. [Online]. Available: https://www.w3.org/TR/wai-aria/ [8] Microsoft, “Locators,” Playwright Documentation. [Online]. Available: https://playwright.dev/docs/locators. Accessed: Aug. 9, 2026. [9] Microsoft, “Auto-waiting and actionability,” Playwright Documentation. [Online]. Available: https://playwright.dev/docs/actionability. Accessed: Aug. 9, 2026. [10] Model Context Protocol Contributors, “Model Context Protocol specification,” 2025. [Online]. Available: https://modelcontextprotocol.io/specification/2025-03-26/index