Skip to content

Starter mental model suggestions

Mental models should start as explicit, inspectable user choices. Pi Hindsight may suggest useful recurring questions, but it must not create or refresh mental models automatically during setup.

Starter suggestions live in docs only. Mental models are created and managed in the Hindsight control-plane web UI; Pi does not expose mental model creation. Use the suggestions below as copy-paste material when creating models in the web UI.

  • Suggestions are explicit opt-in.
  • No suggestion creates a mental model automatically.
  • Creation should show the model name, bank, source query, and tags before submission.
  • Refresh remains explicit until a later issue defines safe refresh policy.
  • Global models must be conservative and avoid transient project details.
  • Suggested source queries should ask for durable patterns, not one-off summaries.
  • Suggested tags should reuse existing project/global visibility language.

Project mental models should describe durable project facts and operating rules.

NameSource querySuggested tagsWhy it helps
Project architecture and seamsWhat are the stable architecture boundaries, modules, and seams in this project?project, architecture, stableHelps future turns understand where changes belong.
Testing and release processWhat test, CI, release, and verification practices does this project use?project, workflow, releaseReduces repeated rediscovery of done criteria and release gates.
Memory safety policyWhat memory rules, safety constraints, and anti-patterns matter in this project?project, memory-policy, safetyKeeps retain/recall behavior aligned with project invariants.
Current roadmap and prioritiesWhat roadmap themes, active priorities, and deferred non-goals recur for this project?project, roadmap, priorityPreserves product direction without turning it into hidden backlog.
Code style and naming conventionsWhat code style, naming, module, and testing conventions recur in this project?project, conventions, styleImproves consistency for future implementation work.

Global mental models should focus on durable collaboration preferences that help across repositories. They should avoid project-specific facts and speculative personality profiles.

NameSource querySuggested tagsWhy it helps
User collaboration preferencesWhat durable preferences has the user shown for collaboration, review, autonomy, and communication?global, user-preference, collaborationHelps agents match working style across projects.
Coding assistant operating preferencesWhat durable preferences has the user shown for how coding assistants should plan, verify, commit, and use tools?global, assistant-workflow, preferenceKeeps agent behavior consistent without copying project-local rules.
Cross-project workflow habitsWhat workflow habits recur across the user’s repositories, issue tracking, PR review, and release process?global, workflow, cross-projectCaptures reusable process patterns without mixing project details.
Tooling and review preferencesWhat tools, checks, review loops, and quality gates does the user repeatedly prefer?global, tooling, reviewHelps pick verification steps and review workflow faster.

Avoid starter mental models for:

  • secrets, credentials, private URLs, or account data
  • single-session status reports
  • speculative personality traits
  • broad autobiographical profiles
  • raw conversation summaries
  • project facts in the global bank
  • global preferences inferred from one isolated project incident

Pi intentionally does not recreate Hindsight’s mental model editor. All create/edit/refresh/delete happens in the Hindsight control-plane web UI. If Pi ever surfaces these suggestions interactively, it must stay read-only with a web interface handoff, and it should not hide source queries behind friendly labels.