Chats

Note

These chats are transformed into Markdown and Notebooks with https://github.com/westurner/transform_md

AgentSession schema

AgentSession:

  • name

  • description

  • dateCreated

  • dateUpdated

  • tags

Agent Session Formats

OpenInference

  • basedOpenTelemetry

agentsview schema

  • kenn-io/agentsview schema (in Go)

Workflow

    1. Research Question

    • “Is it possible to”, “How to”, “Could we/they instead”

    • list of search terms; topics

    • txt, markdown

    • formulate as a sentence or paragraph to ask a search engine / LLM

    • research capture

      • might be captured in an AgentSession or a wiki page,

      • often just an idea

    1. Chat with agent(s), experts(s)

    • context management

      • reuse existing context

      • create new context

        • clean context

          • can attempt to verify that we haven’t talked the model into an ungrounded finding

          • can attempt to verify that another finding was not simply a chance model “hallucination”

        • load the context

          • here, we intentionally bias the model response

          • read precedent information into context first

    • Ask question

Layout:

  • regex: data/chats/*.md

  • regex: data/chats/*.json

  • regex: data/chats/_*.md

    • tool: TODO

  • regex: data/chats/_* .md

    • tool: TODO

  • regex: data/chats/Gemini-_(\d).*

    • tool: TODO

  • regex: data/chats/gemini_google-gemini_.$(date -Is).*

    • tool: TODO

  • regex: data/chats/

Tag Folders TODO

Each of these directories are tag folders; they contain symlinks to other chats:

  find . -type d -maxdepth 1 | sort

  find . -type l -printf '%p'$'\t''%l\n'

chat index.md

We could instead create a YAML-LD metadata document which lists each chat and its metadata fields.

We could instead create a per-chat YAML-LD metadata document which lists each chat and its metadata fields; named like .meta.yml