# Vision and boundaries

## Purpose

thought stream makes Cameron's information environment legible enough for small agents to do continuous, reviewable work. It is a private coordination plane rather than a social dashboard or an autonomous executive.

The system should answer:

- What arrived?
- Where did it come from?
- What changed?
- Which agents noticed it, and why?
- What context did each agent receive?
- What did it infer or propose?
- What model or adapter produced the result?
- Did the run actually complete?
- Can the entire projection be rebuilt from source events?

## Desired behavior

- One stream can contain email metadata, social events, Telegram messages, RSS items, file diffs, timers, git changes, and agent-produced events.
- An agent subscribes to a typed subset instead of receiving the whole personal firehose.
- Cheap triage agents can label, cluster, extract, or route. More expensive agents can be invoked only when a prior result justifies escalation.
- Cameron can watch root activity and inspect complete lineage without being buried in raw trace noise.
- Model behavior can improve through versioned prompts, adapters, and Tinker training while old outputs remain attributable to their exact runtime.
- A conceptualizer consumer can extract concepts and directional links from source events, producing private derived observations that are inspectable and rebuildable without any outbound publication authority.
- A policy-gated Coil consumer can keep one Co conversation per stable document, recommending revisions or additions to Public Knowledge without gaining file-write, staging, publication, protocol, or deployment authority.
- Broad obligations can become governed focus declarations over the same evidence stream, with narrower child focuses proposed rather than silently self-authorized.

## Explicit non-goals for the first release

- A universal autonomous assistant.
- Automatic public posting, email replies, Telegram sends, or file edits.
- Feeding the entire Obsidian vault or inbox into every model call.
- Treating agent summaries as truth.
- Hiding failures behind a green service status.
- Building a rich dashboard before the event loop is real.

## Product test

thought stream is useful when it produces a small number of surprising, inspectable, low-cost observations Cameron would not otherwise have connected, while making it obvious how each observation was produced.