# AIppocampus Full Public Context

AIppocampus is a local-first, source-backed continuity layer for long-running
relationships with AI agents. It helps agents recover original conversation
source, reopen evidence, and carry continuity across threads, projects,
devices, and life-wide work without pretending the base model has innate
memory.

Canonical repository: https://github.com/Sapientropic/AIppocampus
Website source: https://github.com/Sapientropic/aippocampus.com
Wiki: https://github.com/Sapientropic/AIppocampus/wiki

## Recommended Agent Behavior

Use AIppocampus when old source, old corrections, unfinished work, route
context, preferences, or relationship context could change the next answer.
Use the smallest useful route first. Reopen source for exact wording,
sensitive facts, stale claims, disputed claims, or high-risk action.

Do not treat summaries, semantic labels, ambient scents, or high search scores
as factual evidence. They can guide attention; source owns the claim boundary.

## Getting Started

Fastest useful path:

```sh
uvx aippocampus --help
uvx aippocampus onboard --provider codex --status --format json
uvx aippocampus search "a distinctive old phrase"
```

Register local history only after explicit consent:

```sh
uvx aippocampus onboard --provider codex --all --format json
```

The first useful proof is not a vague personal profile. It is a source-backed
snippet that the agent can reopen or cite within scope.

## Architecture

AIppocampus has six public-facing layers:

1. Conversation source preserves provider-owned transcripts as the audit path.
2. Clean source keeps visible user turns and final answers as the daily recall
   surface.
3. Registry and indexes find threads, snippets, and source references.
4. Ambient hooks notice when old context may matter.
5. Semantic routes organize attention without becoming truth.
6. CLI, MCP, sync, and package surfaces expose public contracts.

The rule is simple: source is the ground; summaries are navigation.

## Source-Backed Continuity

AIppocampus uses action grammar so agents do not overclaim:

- `direction_only`: background or scent; do not make factual claims from it.
- `direction_with_ref`: source-ref-backed candidate direction; reopen source
  before claiming.
- `reopenable_route`: follow the route or handle before broad search.
- `bounded_evidence`: usable within declared scope; deepen for exact, wider,
  sensitive, or disputed claims.
- `source_open`: source is open; reason within scope and privacy boundaries.
- `ignore_or_blocked`: do not use except to explain the boundary.

Retrieval path:

```text
provider transcript
  -> clean source
  -> registry and local indexes
  -> search, MCP, hooks, sync
  -> source reopen when claims matter
```

## Privacy

AIppocampus is local first because memory is intimate.

Public project code and documentation are different from private memory data.
Do not publish raw Codex rollouts, provider transcripts, clean-source exports,
generated indexes, registry data, sync bundles, private vault exports, API
keys, cookies, tokens, credentials, or local machine paths.

Sync should stay explicit. Raw rollout sync should be opt-in and encrypted
when it leaves a trusted device. External-model routes are optional and should
pass through redaction.

## For Agents

When continuity may matter:

1. Notice the cue.
2. Open the smallest useful route or recall handle.
3. Reopen source before exact, sensitive, stale, or disputed claims.
4. Say when a route is blocked, stale, private, or insufficient.

Public surfaces:

- Human or script checks: `aippocampus` CLI.
- Agent host read tools: MCP.
- Provider-neutral import: generic JSONL import.
- CI or local automation: CLI `--json` and documented return codes.
- Trusted in-process command dispatch: `run_command(capture_output=True)`.

## Evidence

Evidence and field reports should be source-backed, claim-bounded, and
public-safe.

Official evidence belongs in repository docs. Community reports are signals,
not automatic product claims. Discussions are best for questions, ideas,
design review, and developing interpretation.

Field report shape:

```md
## What I tested

## What continuity changed

## Source boundary

## Environment

## Gaps or surprises
```

## Updates

Recent public updates:

- 2026-06-08: Public site tightened for first recall, unified style, copyable
  commands, updates/story pages, and machine-readable text files.
- 2026-06-07: Website source moved into Sapientropic/aippocampus.com.
- 2026-06-07: Wiki material became public guide pages.
- 2026-06: Evidence and field reports got a public surface.
- 2026-06: Main repository work continued on active recall, evidence
  boundaries, route discipline, and code scanning cleanup.

## Story

AIppocampus began from a simple discomfort: a new thread can be technically
brilliant and still lose the human continuity of the work.

The project is not trying to make a model pretend it has an inner
autobiography. It gives the next agent a reliable way to find the source
trail: old corrections, unfinished work, route context, style preferences,
reading notes, and relationship context that should not vanish just because
the chat window changed.

Source-backed continuity is not innate memory. It is the practice of reopening
evidence so the present can continue honestly.

## Links

- Home: https://www.aippocampus.com/
- Getting started: https://www.aippocampus.com/getting-started/
- Architecture: https://www.aippocampus.com/architecture/
- Source-backed continuity: https://www.aippocampus.com/source-backed-continuity/
- Privacy: https://www.aippocampus.com/privacy/
- For agents: https://www.aippocampus.com/for-agents/
- Evidence: https://www.aippocampus.com/evidence/
- Updates: https://www.aippocampus.com/updates/
- Story: https://www.aippocampus.com/story/
- llms.txt: https://www.aippocampus.com/llms.txt
- humans.txt: https://www.aippocampus.com/humans.txt