Today, we're announcing the Nimble integration for Hermes. With Nimble, developers can build Hermes agents that perform complex web research with higher accuracy using less tokens, giving them better context to reason, decide, and act.

What You Can Do with Web Search Agents + Hermes

Simply send tasks to a Web Search Agent through the API, and it automatically searches, crawls, and extracts the information Hermes needs.

Run over run, the Web Search Agent learns which sources and retrieval paths work best for your domain. This self-learning approach helps it go deeper into the right websites, avoid irrelevant content, and retrieve better context more efficiently over time.

• Improve accuracy. Tailor search and crawling to the agent’s specific task instead of relying on shallow, broadly relevant results from generic search.
• Reduce costs. Reuse learned retrieval paths with cacheing, eliminate unnecessary searches, and return cleaner context that requires fewer tokens to process.
• Automate research workflows. Handle multi-step search, crawling, extraction, and synthesis without developers orchestrating each step.
• Access more of the web. Render JavaScript and navigate dynamic pages that many generic search tools cannot reliably retrieve.
  ‍

‍

Hermes can launch these tasks from conversations or scheduled workflows and retrieve the results later using durable run identifiers, making it easy to build persistent agents that monitor the web and act on reliable findings.

Cookbook: Monitor and Track Company Changes Over Time

This cookbook demonstrates how to turn an company research workflow into a persistent account-monitoring agent (see the code here). The example tracks a watchlist of companies, researches changes on a schedule, records findings in Hermes memory, and sends a concise digest through Slack or another Hermes delivery channel.

Here's how it works:

1. Enable Nimble inside Hermes. Install the hermes-nimble-agent package and enable the plugin. Five Nimble research tools become available to Hermes in conversations, cron jobs, and gateway workflows.
2. Research the watchlist in parallel. The monitor groups accounts into small batches and starts multiple Web Search Agent runs concurrently, improving coverage without requiring Hermes to block while the research is completed.
3. Continue research across scheduled runs. Nimble immediately returns durable identifiers for each research task. Hermes can check those runs during a later cron cycle, allowing research to survive process restarts and span multiple scheduled executions.
4. Build a dated event ledger. Each discovered development is stored in Hermes memory according to the date the event actually happened, together with its account, signal type, source, and confidence.
5. Report only genuine changes. Rather than comparing two non-deterministic research outputs, the monitor filters events by their occurrence date. An older event discovered today is added to the ledger without being incorrectly presented as breaking news.
6. Deliver a cited digest. New developments are sent to Slack, Telegram, Discord, or the terminal, with a citation and confidence grade for each change. When nothing meaningful has happened, the agent stays quiet.
7. Reuse accumulated context. Because the ledger is stored in Hermes memory, users can ask questions about previously researched companies without launching another web research task or incurring additional retrieval costs.
   ‍

‍

The repository includes a sample digest and a prebuilt ledger containing 161 dated, sourced events across 15 companies, so developers can explore the complete workflow before connecting an API key.

Give Hermes Better Web Context with Nimble

Hermes provides the agent loop, memory, scheduling, skills, and messaging infrastructure needed to build persistent agents. Nimble gives those agents accurate, complete, and verifiable context from the live web.

By adapting retrieval to the agent's specific task, reusing what it has already learned, and efficiently extracting structured information, Nimble helps Hermes agents spend fewer searches and tokens finding the context they need. Web Search Agents also handle complex, multi-step research workflows while providing the citations, confidence signals, and auditability required to trust the results.

Together, Nimble and Hermes make it possible to build agents that do more than search the web. They continuously gather better context, retain what they learn, and act on information developers can verify.