Anchor Launches Action Index to Cut AI Agent Token Costs by 100X
New browser infrastructure tool lets AI agents replay proven web workflows as deterministic code, drastically reducing compute costs and execution time.
Key points
- Anchor launched Action Index on October 15, 2026, to share successful web workflow memories among AI agents.
- Testing showed a 100x cost reduction, with full benchmark runs dropping from $472 to $4.87.
- The tool is available immediately to all Anchor accounts, including the free tier, via MCP server and REST API.
Anchor, the browser infrastructure provider for artificial intelligence agents, launched Action Index on October 15, 2026, according to a report by The Next Web. This new feature functions as a shared memory system for web workflows that agents have previously completed successfully, allowing them to bypass redundant reasoning processes.
Instead of analyzing every click independently, agents using Anchor can now replay a proven navigation path as deterministic code, The Next Web reported. The underlying language model only intervenes when a webpage layout or content has changed since the last successful execution, significantly reducing the computational load required for routine tasks.
Performance and Cost Efficiency
In evaluations using 300 tasks from the Online-Mind2Web benchmark, a small model equipped with Action Index completed 93.3% of tasks at an average cost of $0.016 per task, according to The Next Web. By contrast, a frontier model achieved 93% completion at $1.57 per task. The total cost for the 300-task run was $4.87 with the new index, compared to $472 without it.
Without Action Index, the same small model only completed 62.7% of the tasks, The Next Web noted. In a separate paired study involving 120 tasks, the index reduced 90th-percentile token usage by 74%, decreased steps by 75% and cut wall-clock time by 61%. It also improved performance for open-weight models, lifting completion rates from 67.7% to 88.7% for one model and from 58.7% to 84.7% for another.
Availability and Integration
Action Index is available immediately to all Anchor accounts, including those on the free tier, as tools on Anchor’s Model Context Protocol server and through its REST API, according to The Next Web. The system is designed to work with any MCP-compatible agent or model provider, ensuring broad interoperability within the current AI agent ecosystem.
Idan Raman, co-founder and CEO of Anchor, stated that agents currently pay to re-solve problems that are already solved, according to The Next Web. He emphasized that memory, rather than the model itself, is now the primary bottleneck for most web-based work, and that every successful run makes subsequent executions cheaper.
What should you do?
Security and technology professionals should review their current AI agent workflows for repetitive web tasks that could benefit from deterministic replay. Organizations using Anchor should test the new Action Index tools in non-production environments to verify compatibility with their specific model providers and MCP configurations. Teams should monitor token usage metrics before and after implementation to quantify cost savings and performance improvements.
What to do and how to stay safe: Anchor
- Audit existing AI agent workflows to identify repetitive web navigation tasks suitable for deterministic replay.
- Test the Action Index tools in isolated environments to ensure compatibility with your specific model providers and MCP setups.
- Monitor token consumption and execution times to measure the efficiency gains provided by the shared memory system.
- Review Anchor’s documentation for best practices on integrating the index with your current browser infrastructure.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How much does Action Index reduce costs for AI agents?
According to The Next Web, the full run of 300 benchmark tasks cost $4.87 with Action Index, compared to $472 without it, representing a 100x reduction.
Is Action Index available for free accounts?
Yes, The Next Web reported that Action Index is available to all Anchor accounts, including the free tier, via the MCP server and REST API.
How does Action Index improve model performance?
The Next Web stated that the tool lifted completion rates for two leading open-weight models from 67.7% to 88.7% and from 58.7% to 84.7%.
