01
Giving agents a runtime environment
A useful agent needs access to appropriate tools and enough context for its task. My self-hosted Hermes deployment covers the runtime, memory and tool integration. Operational requirements remain central: availability, permissions, secrets and the ability to inspect what ran.
02
Choosing and routing models
Omniroute and my experience with LiteLLM focus on model access and application integration. Routing needs to consider expected quality, latency and budget. Token optimization complements this work by reducing unnecessary context without removing information required to complete a task reliably.
03
Integrating AI into an existing workflow
OpenCode Harness focuses on the working environment of coding agents. M2C Flows applies AI assistance to cloud migration preparation. Both start with an identified task and a verifiable result. Consequential decisions and code approval remain under human control. The project pages explain what was built rather than promising unlimited autonomy.
