How I Give Claude, Hermes, and OpenClaw Live Social Context

How to Provide Context to Your Agents

Introduction and Overview

• The session begins with a welcome message, indicating that the discussion will focus on providing context to agents, specifically using the Hermes agent.

• Participants are encouraged to share their experiences with various tools like Open Claw, Hermes agents, and Claw Code in the chat.

Agent Frameworks: Hermes vs. Others

• The speaker introduces Hermes as an agentic harness framework that is open-source and free, claiming it is superior to Open Claw and Claw Code.

• Concerns about vendor lock-in are discussed; recent degradation of Opus 4.6 model performance highlights the need for flexible systems.

• The speaker uses both Hermes and Open Claw but prefers Hermes for its reliability and better security against prompt injection issues.

Memory Systems in Agents

• A strong memory system is essential for agents; Hermes has an advanced memory capability that outperforms other frameworks.

• The self-learning feature of Hermes allows it to adapt based on user interactions without requiring extensive technical input from users.

Enhancing Contextual Understanding

• To improve agent memory systems, it's crucial to provide structured context; otherwise, interactions may feel disjointed or ineffective.

• Reference is made to Kapathy's philosophy on creating a "Wikipedia" style knowledge base for agents' memories using interconnected files.

Utilizing Obsidian for Knowledge Management

• Obsidian is introduced as a tool for building a knowledge base that links topics together effectively, enhancing recall capabilities of agents.

• The speaker shares how their agent's brain grows daily by integrating new information into this interconnected structure.

Feeding Context into Agents

Internal vs. External Context

• Two types of context are identified: internal (e.g., sales calls, team meetings recorded via Granola or Fireflies) and external (e.g., webinars).

Internal Context

• Internal context includes recordings from team meetings which provide insights into decision-making processes within the organization.

External Context

• External content such as webinar recordings can be valuable; transcripts should be generated from these sessions for integration into the agent’s knowledge base.

Managing Information Flow

• Raw transcripts are processed before being added to the knowledge base to avoid overwhelming the system with unnecessary data.

Automation through Chron Jobs

• Chron jobs automate information processing by updating relevant wiki files while filtering out irrelevant data points.

Practical Applications of Agent Context

Using Help Docs Effectively

• Discussion on how help documentation can be updated automatically based on new features released in applications like Substack.

Training Support Agents

• Up-to-date help docs enhance support agents’ efficiency by reducing manual intervention needed when addressing customer inquiries.

This markdown file summarizes key discussions around providing contextual understanding in AI agents during a live session focused on utilizing tools like Hermes and Obsidian effectively.

Understanding the Role of Social Media in Business Context

The Importance of Context for Agents

• Providing context to agents is crucial, but much of the information they rely on is historical, such as data from Stripe and Superbase.

• Current internal team meetings may offer some future insights, but overall context tends to be siloed and not forward-looking.

Leveraging Social Media for Insights

• Social media platforms serve as a voice of the world, offering real-time insights that can inform business strategies.

• Signals from social media should be viewed as genuine indicators rather than just sales signals; they highlight trends and topics worth exploring.

Identifying Trends Through Signal-Based Platforms

• Platforms like LinkedIn, X (formerly Twitter), Instagram, and Threads are identified as signal-based platforms where trending themes can be monitored.

• Fast-paced content on these platforms allows businesses to stay updated with what audiences care about.

Understanding Layer: Moving Beyond Short Content

• To gain deeper understanding beyond short-form content, longer formats like YouTube videos and podcasts should also be analyzed.

• Tools like Hermes can set up searches across various platforms to monitor relevant discussions and emerging trends.

Building an Effective Agent System

Setting Up Monitoring Systems

• Hermes utilizes different searches within the Triggery platform to track industry-related conversations and intent data.

• The system monitors results for traction before taking action based on viral posts or significant discussions.

Gaining Industry Insights

• By analyzing both trending topics and longer-form content, agents can update their knowledge base effectively.

• This process helps create a comprehensive understanding of industry dynamics relevant to the business's operations.

Performance Evaluation through External Data

• Monitoring external performance metrics helps assess how well the business is perceived in its market space.

• Understanding competitor activities through social data provides valuable insights into product launches and market positioning.

Enhancing AI Capabilities with Contextual Knowledge

Building a Knowledge Layer for Agents

• An effective agent must understand customer pain points and spending habits to provide tailored recommendations.

• A well-developed context layer enables agents to deliver insights that feel personalized rather than generic AI outputs.

Implementing Feedback Loops for Continuous Improvement

• Engaging with feedback mechanisms allows agents to refine their understanding of ideal customer profiles (ICPs).

Practical Implementation Strategies

Tips for Implementing Agent Systems

• Providing prompts is essential for non-coders looking to implement agent systems effectively; clear instructions help streamline setup processes.

Accessing Agents Across Teams

• A Discord channel facilitates communication between team members and various specialized agents designed for different tasks.

Handling Visual Data in Agent Systems

Managing Non-textual Information

• Agents equipped with vision capabilities can interpret visual data formats such as JPEG or MP4 files using APIs that convert audio/visual inputs into text.

Utilizing Campaign Results for Internal Context

Integrating Campaign Data into Agent Learning

• Campaign results can influence agent behavior by providing contextual learning opportunities based on past performance metrics.

Navigating Local Models in AI Development

Exploring Local Model Options

• Testing various local models reveals differences in performance; selecting appropriate models depends on specific use cases within workflows.

Starting Point for New SaaS Companies

Building Initial Knowledge Bases

• New SaaS companies should gather external context from competitors while leveraging direct interactions with customers during initial development phases.

Understanding Agent Utilization in Business

The Need for Multiple Agents

• When tasks are unified across a business, the question arises about the necessity of multiple agents. This is particularly relevant when many team members are simultaneously using the same agent.

• Using a single agent can lead to slow response times and increased usage costs, prompting the need for dedicated agents tailored to specific tasks or models.

Infrastructure and Naming Conventions

• A question was raised regarding infrastructure naming conventions for agents within sequences and campaign data integration. The speaker no longer uses MCPS due to context bloat issues.

• Instead of MCPS, a CLI approach is preferred as it avoids performance degradation associated with large token counts in sessions.

Tracking Campaign Data

• Campaign data tracking involves converting raw files into actionable insights at specified intervals (e.g., 7-day, 30-day cycles). This helps in assessing campaign effectiveness.

• If direct statistics from campaigns cannot be pulled, it's essential to segment campaigns into groups for better hypothesis extraction from campaign data.

Defining Hypotheses and Beliefs

• Establishing clear definitions for what constitutes a created hypothesis or belief is crucial. Different teams may have varying criteria on how often an event must occur before it’s deemed significant.

Resources for Learning and Development

• As the discussion nears its conclusion, participants are encouraged to explore educational resources like YouTube channels that provide valuable content related to agent utilization.

• Recommendations include following influencers such as Alex Finn and Matthew Bman on social media platforms to enhance understanding of industry terminology and practices.

Building Knowledge Before Implementation

• Emphasizing self-education is vital; understanding foundational concepts will aid in effectively building out agents while incorporating best practices.

• Participants are reminded to ensure their systems are well-organized architecturally before implementing new technologies or processes involving agents.

Conclusion and Future Engagement

• The session wraps up with an invitation for feedback on future topics of interest, indicating ongoing engagement opportunities for participants.