AIppocampus

io.github.Sapientropic由 GitHub 帳號 Sapientropic 發布v0.3.3更新於 Oct 24, 2026

Local-first, source-backed continuity for AI agents via stdio MCP.

已驗證STDIO僅桌面知識與記憶開發者工具

概覽

AI 產生的概覽

本地優先的記憶與連續性伺服器,讓助理跨對話、裝置與專案回想、檢索並重新開啟過往對話的原始來源。

可以做什麼
  • 從已登記的對話記憶中回想舊決定、交接、偏好或線索
  • 深化回想路徑,並在下結論或引用前重新開啟原始來源
  • 當使用者記得確切措辭時,對來源做精確檢索
  • 從 Codex 記錄、Claude Code 逐字稿或通用 JSONL 匯出等本機來源建立乾淨來源
  • 匯出與同步記憶產物,讓連續性延續到下一個對話、裝置或專案
適用情境
適合與助理長期進行的編碼或寫作工作,尤其是上下文在對話、裝置或專案之間反覆遺失的情況。適合希望回想以原始對話來源為依據而非僅憑摘要,並願意明確登記本機歷史的使用者。
執行需求
以本機 stdio 程序執行,來自 PyPI 套件 aippocampus,通常以 uvx 啟動;需要 Python 3.12 或更新版本。本機對話來源需經同意後登記。選用設定使用 AIPPOCAMPUS_REGISTRY_DIR、AIPPOCAMPUS_HOME、AIPPOCAMPUS_VAULT 等變數;語意或背景模型工作需 AIPPOCAMPUS_DEEPSEEK_API_KEY 或 AIPPOCAMPUS_OPENAI_COMPAT_* 路由。
安裝前請注意
乾淨來源仍可能包含私人對話文字,原始記錄、登記資料列與 vault 筆記應視為私人歷史。登記本機歷史屬於寫入操作,應僅在使用者明確同意後進行。原始記錄同步在不受信任的多裝置環境中必須先加密。AIPPOCAMPUS_DEEPSEEK_API_KEY 等選用金鑰不得提交或列印,外部模型路由應做去識別化處理。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 AIppocampus,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。

其他 MCP 客戶端

參照 儲存庫 中的啟動說明。

README

[A shadow figure and a light figure clasp hands in a ruined circular hall, with light opening between them.]

AIppocampus

A source-backed continuity layer for long-running relationships with AI agents.

AIppocampus began with a human problem: every new agent session can be bright, capable, and strangely newborn. Work may survive in commits and notes while the path behind the work falls back into silence.

This project gives future agents a way to find that path again. It keeps source reachable, preserves the conditions of return, and lets a new conversation begin with honest continuity instead of pretending there was never a break.

Source is the ground. Summaries are weather.

First Use Path

The first useful path is deliberately narrow:

  1. Install or probe the CLI.
  2. Run one fuzzy recall for an old decision, handoff, preference, or thread cue.
  3. Deepen that route before making claims or quoting source.
  4. Use exact search only when the user remembers wording.

Install or probe:

sh
uvx aippocampus --helpaippocampus start --json

First recall, for a fuzzy continuity cue:

sh
aippocampus agent recall "old decision or handoff cue" --json

First deepen/source-open, using the selector from that recall:

sh
aippocampus agent deepen --request 1 --recall-selector <emitted-selector> --json

Daily rule: if the user remembers exact wording, search source; if they remember a situation, decision, preference, old correction, or handoff, use recall first and deepen before claims.

sh
aippocampus search "a distinctive old phrase" --json

In ordinary use, AIppocampus should feel less like a control panel than a remembered doorway. It helps an agent ask: where did this come from, what did we actually say, and which source should be opened again?

The machinery behind that moment can be extensive, but it should stay backstage. A long relationship with an AI agent should not have to start from bare ground every time a thread, device, model, or project changes.

The origin essay is 未干的地图. English readers can start with The Unfinished Map. For role-based setup, provider choices, and maintainer paths, use Start Here. For a longer first-use walkthrough, use the First Recall Decision Card or the 10-Minute Public Path. Evidence and claim boundaries live after the first route in Magic Moments, Claim-Bounded, the Can-Claim Ladder, and the Public Provenance And Current Value Ledger.

After a useful deepen, aippocampus export --json and aippocampus sync --json are carry-forward checks for the next thread, device, or project. They are not prerequisites for first recall.

AIppocampus For Coding Agents

The narrow usable product lane today is local, source-backed continuity for Codex/Claude-style long-running work: register local conversation source after consent, search or reopen old snippets, and treat recall scents or repo familiarity as navigation until source is reopened.

Start with the Coding-Agent Memory Lane when deciding whether to recommend AIppocampus as an agent-memory tool. It includes the 3-5 minute demo path, evidence drawer, and "recommend / do not recommend" boundary without duplicating the public API or readiness docs.

Agent Probe

When an AI agent needs to verify the public CLI without cloning or writing local memory artifacts, use the PyPI package and stay read-only:

sh
uvx aippocampus --helpuvx aippocampus onboard --provider auto --status --json

This is a provider matrix, not consent to ingest every detected provider. Only after the user explicitly agrees to register local history, route to Start Here or the First Recall Decision Card for the right provider-specific write path.

sh
uvx aippocampus search "without pretending it has innate memory" --clean-source-dir ./examples/public-memory-bundle/clean-source --json

That exact-search demo uses bundled public clean source and does not touch private history. Agents should read docs/agent-context.md and llms.txt before recommending or comparing AIppocampus. Use the GitHub uvx --from git+... form only when intentionally testing an unreleased main-branch snapshot.

What It Is

AIppocampus is the canonical public home for an external hippocampus for agents: a local-first memory and continuity system that keeps original conversation source available across projects, devices, and casual life-wide conversations. The hippocampus language is a design metaphor, not a biological claim; the architecture overview maps each major metaphor to its runtime mechanism, current claim, and boundary.

Project work is one surface. The deeper aim is continuity across work, reading, reflection, unfinished questions, and the small phrases that make a relationship recognizable again.

What It Carries

At the center, AIppocampus keeps source close enough that continuity can be honest instead of theatrical:

  • Builds clean source from supported local conversation providers: Codex rollouts, Claude Code transcripts, or explicit generic JSONL visible-message exports, with summaries kept as navigation layers.
  • Searches old conversation memory across the current thread and registered threads, then treats exact hits as source-backed snippets.
  • Reopens or cites source when the agent needs evidence instead of memory-like impressions.
  • Stores generated recall artifacts in the configured AIppocampus registry (AIPPOCAMPUS_REGISTRY_DIR, AIPPOCAMPUS_HOME/registry, then legacy $CODEX_HOME/aippocampus-registry) so memory remains useful when a new project opens. Project-local .aippocampus/ output is explicit compatibility or export mode.

Ambient hooks are close to the front door because they keep continuity from collapsing back into manual search. There are still deeper doors for people who want them: MCP wiring, sync, plugin packaging, diagnostics, review surfaces, semantic workers, and research experiments. They matter, but they should not stand in front of the first handshake: source found, source reopened, continuity resumed.

First Stops

Reading For The Soul

The research notes carry the human shape of the project. They are speculative frames, not runtime contracts, but they explain the taste behind the machinery:

  • The Pearl of Presence asks why retrieval without accumulated acquaintance can still feel absent.
  • Source as World, Interpretation as Weather gives AIppocampus its grounding rule: many meanings can grow from one shared world, and the world must have happened.
  • Journey Tracking follows continuity as a first-person plural journey, with source-backed waypoints instead of a flat user profile.
  • Dream Task Design sketches the subconscious layer: quiet work that integrates what the foreground could not finish.
  • Ambient Associative Recall describes how old memory can return as a scent before it becomes an interruption.
  • Long Garden keeps far-future seeds without turning them into default product promises or open-issue clutter.

Operator And Maintainer Paths

The README stays close to the product path. Operator commands, source-checkout setup, plugin packaging, sync, release checks, and benchmark notes live in the guides that own those contracts:

AIppocampus supports Python 3.12 and newer. On macOS, Homebrew Python 3.12 is the documented baseline before running source-checkout verification.

For repository contributors, the dev extra install path is:

sh
python3 -m pip install -e ".[dev]"

Public claims still need the maintainer lanes in CONTRIBUTING.md. The default CI keeps one canonical Ubuntu Python 3.12 pr lane with coverage, a Python 3.13 quick compatibility lane, and a focused macOS default TMPDIR path-identity gate; Ubuntu green alone is not a broad cross-platform path-identity claim. The broader boundary lives in docs/architecture/source/path-identity.md.

Privacy Boundary

AIppocampus is local-first.

  • Clean source may still contain private conversation text.
  • Raw rollouts, bundles, registry rows, vault notes, and generated archives should be treated as private history.
  • External-model routes are optional and should use redaction safeguards.
  • Raw rollout sync should stay explicit and must be encrypted before use with untrusted multi-device sync. Start with the read-only Sync Decision Card before choosing push, pull, repair, or object storage.
  • Do not commit personal rollouts, .aippocampus/ outputs, registry data, API keys, cookies, tokens, or private vault exports.

Common non-secret configuration:

  • AIPPOCAMPUS_REGISTRY_DIR / AIPPOCAMPUS_HOME
  • AIPPOCAMPUS_VAULT
  • AIPPOCAMPUS_STYLE_SOURCE
  • AIPPOCAMPUS_SCRIPT_SOURCE
  • AIPPOCAMPUS_SITE_MARK
  • AIPPOCAMPUS_SITE_TITLE
  • AIPPOCAMPUS_SEMANTIC_GATE

Optional provider secrets are separate. Basic source search, MCP/plugin setup, and local hooks do not require them. Set AIPPOCAMPUS_DEEPSEEK_API_KEY or an AIPPOCAMPUS_OPENAI_COMPAT_* route only when you explicitly want semantic or background model work. Provider-native env names are custom-route choices, not built-in defaults. Values must never be committed or printed.

Roadmap

The root roadmap pointer is ROADMAP.md. The canonical detailed roadmap lives at docs/roadmap.md. The documentation map is docs/README.md.

Repository Layout

text
AIppocampus/|- skills/aippocampus/        # installable skill package|- skills/aippocampus-ux/     # agent-facing UX review skill|- plugins/aippocampus/       # Codex plugin source package|- docs/                      # origin essay, design notes, guides, evidence|- docs/guides/assets/        # public README and documentation artwork|- sources/                   # lightweight provenance catalog|- tests/                     # repository-level unit and integration tests|- tools/                     # smoke, docs-health, and maintenance tools|- README.md|- ROADMAP.md|- AGENTS.md`- LICENSE

License

The public AIppocampus repository is licensed under Apache-2.0.

The Apache-2.0 public core covers the code, docs, local tools, schemas, MCP surface, plugin packaging, public examples, and bundled project artwork shipped in this repository unless a bundled third-party asset says otherwise. Hosted services, enterprise governance, managed graph/semantic layers, support, and other operated product surfaces can be offered under separate commercial or product-specific terms.

Private user memory data is not project code. Raw rollouts, clean-source exports, registry rows, sync bundles, vault exports, generated indexes, and thread anchors remain private user artifacts unless their owner explicitly publishes them.

See docs/guides/public-core-boundary.md for the canonical licensing, adapter, schema, third-party asset, and relicensing boundary. See docs/guides/public-api.md for supported CLI, MCP, environment-variable, JSON, and import-stability expectations.

來源:README.md,提交 5327a3d

工具

0
SourceWeft 尚未在沙箱中啟動過該軟體包,因此其工具未知。

版本歷史

1
  1. v0.3.3最新Sep 16, 2026

評分

已滿足適用檢查項的 50%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通25 分

    請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。

  • 已宣告認證方式10 分

    請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。

  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分