Uploaded July 2026 | Updated September 2026, 2 weeks ago
Hermes Agent is Nous Research’s open source, command-line-first AI agent framework built to run locally, connect to your tools, remember workflows, and grow with the user over time.
In this talk, Sam Herring, machine learning engineer at Nous Research, shares the story behind Hermes Agent: how it began as an internal productivity tool, why Nous rebuilt its own agent stack in Python, and how the team designed it around open source, open weights, local execution, private data, long-running sessions, and tiered memory.
Sam walks through Hermes Agent’s core architecture, including its agent loop, 80+ tools, 80+ bundled skills, automatic skill writing, messaging integrations, scheduled tasks, memory providers, prompt caching, local model support, and provider flexibility. He also shares personal examples of using Hermes Agent for visual search, production coding, research experiments, reinforcement learning environments, and early recursive self-improvement workflows.
The talk closes with examples from the Hermes Agent community, including music and video generation, smart home automation, research assistants, hackathons, and ways to install, customize, and contribute to Hermes Agent.
Chapters:
00:00 Intro: Sam Herring from Nous Research
00:25 What is Hermes Agent?
01:00 Why Nous built an OpenCode-style agent
01:44 From internal tool to open source product
02:41 Local-first agents, open weights, and private data
03:24 Tools vs skills in Hermes Agent
04:30 Automatic skill writing and workflow memory
05:25 Messaging integrations and async agent workflows
06:17 Scheduled tasks and Sam’s car-buying example
07:22 How Hermes Agent was built
08:04 The classic agent loop
08:40 Compaction and tiered memory providers
09:38 Skills, prompt caching, and repeatable behavior
10:31 Extending Hermes Agent by asking it to modify itself
11:25 Model providers, local hosting, and open access
12:29 Building visual search with Hermes Agent
14:14 Using Hermes Agent as a daily coding agent
15:05 Running research experiments asynchronously
16:12 Reinforcement learning and recursive self-improvement
17:29 Community skills and creative applications
18:25 Home automation and local agent use cases
19:22 Research assistants and literature review workflows
20:18 Hackathons, free trials, and supported models
21:15 Getting started with Hermes Agent
21:45 Contributing to the Hermes Agent community
Subscribe to Arize for more talks on AI agents, open source AI, LLM evaluation, AI observability, agent frameworks, and production AI systems.
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
Hermes Agent is Nous Research’s open source, command-line-first AI agent framework built to run locally, connect to your tools, remember workflows, and grow with the user over time.
In this talk, Sam Herring, machine learning engineer at Nous Research, shares the story behind Hermes Agent: how it began as an internal productivity tool, why Nous rebuilt its own agent stack in Python, and how the team designed it around open source, open weights, local execution, private data, long-running sessions, and tiered memory.
Sam walks through Hermes Agent’s core architecture, including its agent loop, 80+ tools, 80+ bundled skills, automatic skill writing, messaging integrations, scheduled tasks, memory providers, prompt caching, local model support, and provider flexibility. He also shares personal examples of using Hermes Agent for visual search, production coding, research experiments, reinforcement learning environments, and early recursive self-improvement workflows.
The talk closes with examples from the Hermes Agent community, including music and video generation, smart home automation, research assistants, hackathons, and ways to install, customize, and contribute to Hermes Agent.
Chapters:
00:00 Intro: Sam Herring from Nous Research
00:25 What is Hermes Agent?
01:00 Why Nous built an OpenCode-style agent
01:44 From internal tool to open source product
02:41 Local-first agents, open weights, and private data
03:24 Tools vs skills in Hermes Agent
04:30 Automatic skill writing and workflow memory
05:25 Messaging integrations and async agent workflows
06:17 Scheduled tasks and Sam’s car-buying example
07:22 How Hermes Agent was built
08:04 The classic agent loop
08:40 Compaction and tiered memory providers
09:38 Skills, prompt caching, and repeatable behavior
10:31 Extending Hermes Agent by asking it to modify itself
11:25 Model providers, local hosting, and open access
12:29 Building visual search with Hermes Agent
14:14 Using Hermes Agent as a daily coding agent
15:05 Running research experiments asynchronously
16:12 Reinforcement learning and recursive self-improvement
17:29 Community skills and creative applications
18:25 Home automation and local agent use cases
19:22 Research assistants and literature review workflows
20:18 Hackathons, free trials, and supported models
21:15 Getting started with Hermes Agent
21:45 Contributing to the Hermes Agent community
Subscribe to Arize for more talks on AI agents, open source AI, LLM evaluation, AI observability, agent frameworks, and production AI systems.
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
