Anchor Launches Action Index, a Shared Memory of Web Workflows for Computer-Use Agents

Anchor, the browser infrastructure company for AI agents, has launched Action Index — a shared memory of web workflows that agents have already completed successfully — the company said on 15 October 2026.

Instead of reasoning through every click from scratch, agents on Anchor can replay a proven path as deterministic code, with the model stepping in only where a page has changed, according to the company. Action Index is available today to all Anchor accounts, including the free tier, as tools on Anchor’s MCP server and through a REST API, and is said to work with any MCP-compatible agent or model provider.

How Anchor describes the product

Computer-use agents that drive a real browser often burn tokens re-discovering the same checkout flow, form path or settings page. Action Index is Anchor’s answer: store successful workflows as reusable execution memory, then replay them unless the site has shifted.

Idan Raman, co-founder and chief executive, said: “Agents today pay to re-solve problems that are already solved. The model isn’t the bottleneck for most web work anymore. Memory is. Action Index means the hundredth run of a workflow costs a fraction of the first, and every successful run makes the next one cheaper.”

Anchor’s own evaluation numbers

The GlobeNewswire headline frames the launch as cutting computer-use agent token costs “by 100x.” That 100x line is Anchor’s own framing in the release title, not an independent audit.

In Anchor’s evaluation on 300 tasks from the Online-Mind2Web benchmark, the company said:

• A small model with Action Index completed 93.3% of tasks at an average cost of $0.016 per task.
• A frontier model completed 93% at $1.57 per task.
• The full 300-task run cost $4.87 with Action Index, compared with $472 for the frontier-model run as presented.
• Without Action Index, the same small model completed 62.7%.

Anchor did not name the small model or the frontier model in the release.

In a separate paired study on 120 tasks, Anchor said Action Index reduced 90th-percentile token use by 74%, steps by 75% and wall-clock time by 61%. Two leading open-weight models — again unnamed — rose from 67.7% to 88.7% and from 58.7% to 84.7% success, per the company.

TSN has not independently verified these figures. They should be read as Anchor’s self-reported results on its chosen task sets.

Availability

Action Index is live for all Anchor accounts, including free tier, exposed as MCP server tools (https://api.anchorbrowser.io/v1/mcp in the release) and via REST, with documentation pointed at Anchor’s MCP docs. Because it sits behind MCP, other agent frameworks that speak the protocol can call it without being locked to one closed model vendor — at least as Anchor describes the integration surface.

Why it matters

Token cost and latency dominate the economics of agents that click through the open web. If replaying a stored workflow can match a frontier model’s success rate at a fraction of the cost — as Anchor claims in its 300-task comparison — then memory and caching become as strategic as model choice for production automations.

The caveats matter equally: unnamed models, company-run benchmarks, and a headline “100x” that compresses a specific cost comparison into marketing language. Sites that change layout frequently will still force the model back into the loop, which is exactly when Action Index is supposed to yield control again.

What’s still unknown

• Which models were used in the “small,” “frontier” and open-weight tests.
• How Action Index behaves on sites with aggressive anti-bot measures or frequent A/B tests.
• Retention, sharing and privacy rules when workflows from one account enrich the shared index.
• Pricing for heavy production use beyond the free-tier availability claim.

What to watch

Watch for named-model reproductions by third parties on Online-Mind2Web or similar suites, and for Anchor to publish which models sit behind the headline comparison.

What this means

• Launch is live. Action Index is available now, including on Anchor’s free tier via MCP and REST.
• Benchmarks are Anchor’s. Success rates, dollar costs and the “100x” framing are company claims.
• Models are unnamed. The release does not identify the small, frontier or open-weight models tested.

Sources

1. Anchor, “Anchor Launches Action Index, Cutting Computer-Use Agent Token Costs by 100x”, GlobeNewswire, 15 October 2026

Related stories

• AWS and Anchor Browser Let AI Agents Pay for x402-Protected Web Pages and APIs on Their Own
• Pine AI Launches Pine Computer, a Sandboxed Cloud Computer for AI Agents, in Private Beta
• Claude Haiku 5.5: Anthropic’s Small Model Gets Much Cheaper and Much Better at Using a Computer