AI Agents and Hermes, from the inside
40 episodes 109 min free no account
Forty episodes, roughly three minutes each. No diagrams pretending to be code: real payloads, real tool calls, real failures.
play --course hermes --resume
ready
tree /course --depth=1
The four parts
4 parts 40 episodes in order
Chapter I E01–E07 07
Foundations
What an agent is, what a model is not, tokens, messages, sampling.
Chapter II E08–E16 09
Tool calling
The wire format, schemas, failures, the loop, and where it stops.
Chapter III E17–E26 10
Context and cost
Why context degrades, compression, memory, identity, permissions, price per turn.
Chapter IV E27–E40 14
Hermes, from the inside
One real runtime, component by component, ending on a traced turn.
- E27 Hermes: The Big Picture 2:48
- E28 Hermes: Ingress 2:44
- E29 Hermes: The Turn Lease 2:34
- E30 Hermes: The Conversation Loop 2:50
- E31 Hermes: The Tool Registry 2:33
- E32 Hermes: Approvals In Practice 2:29
- E33 Hermes: Context Compression 2:40
- E34 Hermes: Prompt And Cache 2:47
- E35 Hermes: Failure And Failover 2:50
- E36 Hermes: Credentials 2:54
- E37 Hermes: State And Search 2:54
- E38 Hermes: Skills 2:31
- E39 Hermes: Background Review 2:44
- E40 Hermes: Trace One Turn 2:53
now—part two
The dissection: Hermes, line by line
Part one explains the architecture. Part two opens the code: 64 episodes across 8 parts, following real files in a running agent. Each episode ends on the exact function it dissected.
59 published · 5 still in production
in production
#—episodes