We want the Agents to be able to natively contribute to HM content.
User Stories
Translation
Summarization
Finding links
Critical Analysis
Idea generation
Importing knowledge
Community pays for tokens? Individuals pay for tokens?
Hide the use of my AI, keep context private
Publish the context and use of my AI, model metadata
Workflows
External Agent interface?
Claude Website
Skills with Claude Code
Internal Agent UX
Definitions
Agent
What is it for us?
System Prompt: Soul / Goal
Model Provider: Provider URL, API key, specify which model
Tools: what can the agent do
Which account, which types of mutations (read only? comment only? everything)
External tools like Web Search + Scrape
Access to other agents (compaction agent)
Schedule and Triggers: when does the model kick into action
Who owns an "agent"
Owned by a node? Because somebody needs to actually run the agent
Node has access to keys, private API keys
Needs some kind of administrator account or a node operator
Context
A Single chat history of an agent
Authorship
Sometimes the human wants to take responsibility for the agent's work. Sometimes the human wants to hide the use of AI and wants to take the credit from the model.
Other times, the human wants to provide full context about the creation of the content. Attribution to the agent, metadata that links to the context where the agent was prompted, and metadata about the model/time/cost/server.
hm://z6MkgisVMELvqnsCo3dYmtVpy8PiqPGMVwfAyBWFn84vebq4/:profile?v=bafy2bzaceciguevcgg7lacbo7fgix52zw3ewydimhmvbgyfzepmliapuaepam&l believes that both workflows are valid, but we should encourage giving the agent authorship, with the context and metadata shared transparently.
Option A: Per-Human-Account Agent Authors
asdf
Option B: Per-Agent Authors
A different author for each combination of
Option C: Human Authorship
Agents will use the human identities to sign.
Distributed Objects
Sync Agents and Context using our universal resources API, just like we want to do with drafts.
This allows collaboration between users in a community and syncing between device.
Principles
Flexibility
We want to provide the most powerful context management system as possible
Do you like what you are reading? Subscribe to receive updates.
Unsubscribe anytime