Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Agent Historian

Agent Historian is a reusable architecture and integration project for cross-session agent continuity.

It separates three concerns that are often conflated:

Agent Bookkeeper
  -> capture, canonical session evidence, archive indexing, search, and provenance

Optional learned-context workflow
  -> deliberate, agent-authored, evidence-backed project briefings

The project is intentionally not a replacement for source control, repository documentation, or human judgement. Current code, repository instructions, and decision records remain authoritative over learned context.

Status

Archive/search is the accepted first product boundary. The Hindsight transcript mining pilot established useful limits, but is not an adopted integration: transcript-only extraction cannot authoritatively model a mutable codebase. Any later learned-context workflow is separate from Bookkeeper and is fed by deliberate agent curation over current project state and source evidence.

Principles

  • Preserve raw session evidence before deriving indexes or learned summaries.
  • Keep archive/search and learned memory independently rebuildable.
  • Make source revision and provenance visible to every derived result.
  • Treat the current checkout, repository documentation, and decision records as authoritative; learned context is supporting, revision-scoped interpretation.
  • Fail open: a memory-system outage must not prevent an agent from working.
  • Keep transport, storage, and deployment configuration operator supplied.

Documents

The companion Agent Bookkeeper project is the concrete capture, archive, and search product.

Validation

Run:

rtk git diff --check

About

Personal agent continuity architecture: evidence archive, retrieval, and learned context.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors