Skip to main content
Glama

Aelios

A long-term memory for your AI. Switch windows, switch clients, and switch models while keeping one continuous memory.

English | 简体中文 · QQ group 1091783659

Aelios runs on Cloudflare Workers as a memory gateway. Your Chatbox, Claude Code, Codex, or custom agent connects to Aelios, and Aelios forwards the request to your model while recording conversations and bringing back relevant context on every turn.

your client  →  Aelios (identify, recall, record)  →  upstream model

Let an AI set it up

If you use Claude Code, Codex, Cursor, or an AI connected to the Cloudflare MCP server, send it this prompt.

Install and set up Aelios (my AI memory gateway) for me. Read https://raw.githubusercontent.com/Aelios-project/Aelios/main/docs/setup-with-ai.md first and follow it. When I need to click something in the browser or enter a key myself, stop and tell me where.

This works from a blank Cloudflare account or an existing deployment. The AI forks and deploys Aelios, creates the Cloudflare API token, configures your first assistant, and optionally imports old memories or chat logs, then hands you a Base URL, API Key, and model name. You only need a free Cloudflare account and a model provider key. You approve login in the browser and store provider keys in your Cloudflare dashboard so credentials stay out of chat. The guide is written in Chinese, and the AI will talk to you in your language.

The recommended setup pairs API mode with the MCP connector. API mode records and recalls automatically on every turn, while MCP lets the assistant actively search, remember, and manage its own notes. If you only have a Claude subscription without API access, connect MCP on its own (and add the Claude Code hook if you use Claude Code).

To install manually, follow the three steps below.

Related MCP server: second-brain-cloudflare

Quick start

1. Deploy

Fork this repository first, then deploy from your fork so future updates stay one click away (see Updating to the latest version).

  1. Click Fork at the top right of this page.

  2. In the Cloudflare dashboard, open Workers & Pages → Create → Import a repository, connect GitHub, and select your fork. Keep the project name companion-memory-proxy, set the build command to npm ci, and set the deploy command to npm run deploy. The deploy script creates the D1 database, Vectorize index, and Queue automatically and applies all migrations.

  3. Open your Worker's Settings → Variables and Secrets and add a Secret named CHATBOX_API_KEY with a strong random password (for example openssl rand -base64 32). This is the only required secret. Optional keys are documented in SECRETS.md.

After deployment, Cloudflare gives you a Worker URL like the following.

https://companion-memory-proxy.<your-subdomain>.workers.dev

Always deploy through npm run deploy. Running a bare wrangler deploy skips resource provisioning and migrations.

One-click deploy button Deploy to Cloudflare

The button creates an independent copy in your GitHub account without a Sync fork button, which makes future upgrades manual. If you still deploy with it, fill in CHATBOX_API_KEY, keep Vectorize Dimensions 1024 and Metric cosine, use build command npm ci, and use deploy command npm run deploy. See Deployed with the button to switch an existing button deployment over to a fork.

2. Add an assistant

Open your admin panel in a browser.

https://<your-worker-url>/admin

Sign in with your Worker URL and CHATBOX_API_KEY, then open Settings.

  1. Under Models & upstream, enter your 32-character Cloudflare account ID (or a custom OpenAI-compatible base URL). For full chat forwarding through Cloudflare AI Gateway, also add CLOUDFLARE_API_TOKEN as a Worker Secret.

  2. Under Assistants, click Add assistant and fill in the fields.

    Field

    Purpose

    Example

    Name (slug)

    URL segment in lowercase letters

    coder

    Primary models

    Only matching models recall and record memory (supports *)

    anthropic/claude-opus-4-5, *deepseek*

    Keys

    Which Worker API keys may use this assistant

    check Main key (CHATBOX_API_KEY)

  3. Click Save.

Each assistant gets its own URL path. An assistant named coder is served at the following base URL.

https://<your-worker-url>/coder/v1

Only registered primary models trigger recall and recording. Utility models such as background classifiers or lightweight sub-models pass through quietly.

When you use coding agents such as Claude Code, Codex, Cline, or opencode, Aelios records spoken conversation only. Code blocks are stored as [code], and shell or file-editing tool steps are omitted. To record everything verbatim, turn off the coding speech filter in the assistant's advanced options.

3. Connect your client

Use CHATBOX_API_KEY as the API key and include the provider prefix in the model name (for example anthropic/claude-opus-4-5 or deepseek/deepseek-chat).

Client

Configuration

Chatbox, Cherry Studio, OpenAI-compatible apps

Base URL https://<your-worker-url>/coder/v1

Claude Code

ANTHROPIC_BASE_URL=https://<your-worker-url>/coder

Codex

base_url = "https://<your-worker-url>/coder/v1" with wire_api = "responses"

Calling /v1 without an assistant slug routes to the first assistant authorized for that API key.

Test the loop by sending "Please remember my test codeword is apple-star-0428.", then asking "What is my test codeword?" in a fresh conversation.

Updating to the latest version

GitHub forks stay on the commit where you forked them until you sync.

  1. Open your fork on GitHub and click Sync fork → Update branch.

  2. Cloudflare detects the new commit, builds, runs npm run deploy, applies any new D1 migrations, and deploys automatically.

Your memories, settings, assistants, and secrets live inside your Cloudflare account. Keep runtime configuration inside /admin so syncing your fork never hits merge conflicts.

Deployed with the button

Switching from a button-created copy to a fork keeps all existing Cloudflare data intact.

  1. Fork Aelios-project/Aelios on GitHub.

  2. If you renamed the D1 database or Vectorize index during the button setup, add CMP_D1_NAME and CMP_VECTORIZE_NAME under Settings → Build → Build Variables and Secrets with your custom names so the deploy script reuses them.

  3. Under Settings → Build, click Disconnect, then Connect and select your fork with build command npm ci and deploy command npm run deploy.

  4. Merge any pull request Cloudflare opens on your fork to align the Worker name.

How Aelios works

  • Per-turn recall On every human turn to a primary model, Aelios searches Vectorize (memo-kb) and D1 full-text indexes, scores passages in one batch with Workers AI reranker (@cf/baai/bge-reranker-base), and appends the relevant clean originals to that turn. No generative LLM runs on the recall hot path.

  • Daytime Dream Enabled by default (ROLLING_DREAM). As new chats accumulate (10 messages or 8,000 characters by default), @cf/cloudflare/clef-flash checks whether the stretch contains anything worth keeping, and DREAM_MODEL (default workers-ai/@cf/zai-org/glm-4.7-flash) queues review candidates. Repeated explicit decisions on the same subject share one candidate for the day. Nightly review sees the last confirmation after the day's extraction is complete.

  • Nightly Dream and Clef review A daily cron (10 20 * * * UTC) processes the previous day's remaining chats, reviews pending candidates with @cf/cloudflare/clef (CLEF_AUTO_REVIEW, on by default and undoable per item), and writes the daily journal. It uses the remaining time to catch up older dates within a three-day window. The cron also handles weekly and monthly journals and retention.

  • Multi-space reading Each assistant writes to one namespace and can read up to 8 spaces via readNamespaces (for example reading an archived space or shared documentation alongside its own space).

  • Provenance and cleanup Raw chat lines live for 7 days by default (MESSAGES_RETENTION_DAYS). Unused memories expire after 180 days (pinned, identity, and persona memories never expire) and are hard-deleted 30 days later. The trash button on each line in /admin offers Delete this line, which keeps memories made from it, or Delete with its memories, which also removes memories that explicitly cited that line.

For the full storage, Dream, provenance, and current-decision architecture, read docs/memory-lifecycle.md.

Daily management in /admin

Open /admin to inspect and manage every memory space. The sidebar switches between Chinese and English (following your browser language by default) and between light and dark themes.

Page

Purpose

Home

Today's conversations and the latest Daytime Dream result. Last night’s card shows chat processing, candidate review, and journal writing for the same dialogue date

Review

Pending memory candidates. Clef classifies unfamiliar Daytime Dream operations during its nightly evidence review. Proposals it cannot resolve stay visible with a reason. Memory changes appear under Auto-handled this week with one-click undo. When Clef is off, candidates wait for manual or MCP review; an unclassified proposal can be kept as a new memory or discarded. Imported candidates always wait under Imported

Journal

Daily, weekly, and monthly entries (weeks older than ~5 weeks fold into the monthly entry), with Rewrite from the originals for days still within transcript retention

Memories

Browse, search, edit, merge, and inspect version history across Core, Keepsakes, Our words, Notebook, Handoff, and World facts

Star map

Interactive 3D memory graph opened in its own page (/admin/starmap)

Dream activity

Daytime retries, nightly batch history, and actual memory changes by date. Open technical details for model names and run IDs. Continue a date from its saved progress

Import & export

Backup export (md or jsonl), structured entry import, chat log extraction, and zero-LLM chat archiving

Settings

Common settings, assistants, models & upstream (including ChatGPT plan and model list), all settings, keys, and troubleshooting (recall excerpts with original messages and vector maintenance)

Set First met under Settings → Assistants to see how many days you’ve known the selected assistant on Home. The card uses your browser’s calendar date. A brief shower of ✨ marks day 520, each 100-day milestone, and each anniversary of the date you met (Feb 29 falls on Feb 28 in other years). On those days the gateway also tells the assistant, in that memory space's first three messages of the day by DREAM_TIME_ZONE, so it can bring it up. Coding sessions are left out.

Running Dream by hand

To catch up a missed night or run a specific date right away, follow these steps.

  1. Open Dream activity and check the memory space under Viewing.

  2. Choose a chat date under Continue by date. Continue processing on Home fills in the date and turns off preview for you.

  3. Turn off Preview only if it is selected, then click Continue. The background task picks up from saved progress. You can close the page once it is queued.

  4. Each run handles one batch (DREAM_MAX_MESSAGES, default 40). Repeat if more chats remain. If the date's extraction is already complete, continuing can run review and write a missing journal without extracting those chats again.

If a task remains unconfirmed for 20 minutes and you have confirmed it stopped, click Mark as stopped to continue from its saved progress.

Export, import, and chat archive

Everything on Import & export operates on the currently selected space.

  • Export Downloads md or jsonl. Markdown puts one memory per line with date, content, #tags, and a backtick key. Editing and re-importing updates matching keys in place while preserving revision history.

  • Import entries Accepts lines like - 2025-12-15 [preference] Likes oolong tea #drinks (date, type, and tags are optional) as well as plain lists or paragraphs. You can save directly (exact duplicates skip, close matches route to Review) or send everything to Review.

  • Import chat logs Parses ChatGPT and Claude conversations.json, SillyTavern .jsonl, and Name: message transcripts in the browser, chunks by day, and extracts candidates via DREAM_MODEL into Review → Imported. Only lines cited by candidates are stored alongside them, independent of 7-day raw chat cleanup. Re-selecting the same file resumes where you left off.

  • Chat archive Selecting Archive originals only for keyword search saves full transcripts with zero model calls. Turn on Archive new chats (CHAT_ARCHIVE_ENABLED) under Settings → Data retention to archive new conversations automatically. MCP clients search archived text with search_chat (2 to 120 characters) and pass a hit ID to search_chat's around parameter to view nearby messages in the same conversation. Archives stay indefinitely by default (CHAT_ARCHIVE_RETENTION_DAYS).

Integrations and optional features

Full Cloudflare AI Gateway and custom model list

  1. In Cloudflare AI → AI Gateway, create a gateway and store your provider keys (BYOK).

  2. Add CLOUDFLARE_API_TOKEN as a Worker Secret (permissions AI Gateway Run, AI Gateway Read, Workers AI Read, and optionally Vectorize Read).

  3. Enter your 32-character Cloudflare account ID under /admin → Settings → Models & upstream, then configure your assistants.

Without AI Gateway or a Cloudflare token, memory recall, MCP, and Workers AI Dream still run out of the box.

To keep client model pickers uncluttered, open Settings → Models & upstream → Model list → Edit to pick the models you actually use, add custom provider or dynamic-route names, and test connectivity per model. GET /v1/models then returns that exact list in order while keeping all other models callable by name.

Use your ChatGPT Plus or Pro plan

Connect your personal ChatGPT plan via OpenAI's Sign in with ChatGPT without an OpenAI API key.

  1. Open /admin → Settings → Models & upstream → ChatGPT plan → Manage → Continue with ChatGPT and sign in.

  2. Your browser redirects to an unreachable 127.0.0.1 URL. Copy that full address from the address bar, paste it back into /admin, and click Finish sign-in.

  3. Call models as chatgpt/<model> (click Test to inspect available models) and add chatgpt/* to an assistant's primary models to enable memory.

Tokens are encrypted in D1 with a key derived from CHATBOX_API_KEY (rotating that key requires signing in again). Chat Completions, Anthropic Messages, and Responses clients all work on this route, including function tools and web search, while hosted tools (image generation, file search, code interpreter, remote MCP) are unsupported. You can also set DREAM_MODEL or DIARY_MODEL to chatgpt/<model>, which counts toward your plan usage. Technical details on GPT prompt caching and recall placement live in docs/memory-gateway.md.

MCP connector (Claude app, Claude Code, Codex)

Connect any remote Streamable HTTP MCP client to the following URL.

https://<your-worker-url>/mcp?token=<CHATBOX_API_KEY>

Prefer MEMORY_MCP_API_KEY when sharing connector URLs across devices because ?token= is stored by clients.

  • Claude web and mobile apps Add a custom connector at claude.ai/customize/connectors with the URL above and set read-only tools to Always allow. Because the official Claude app ignores server-supplied MCP instructions, paste the guidance below into your user preferences or project instructions.

You have access to my Aelios memory connector for long-term memory. Call wake_up at the start of every new conversation before replying. Call recall whenever I mention people, past events, or preferences, or before guessing something about me. Call remember right away when I share new facts, change my mind, or make a promise. Call log_conversation as the chat winds down so tonight's diary and memory pass can reflect on it.
  • Direct-chat clients Chats in the official Claude app skip the gateway, so log_conversation hands the transcript to Aelios for daytime/nightly Dream and journals. Gateway clients record automatically and skip log_conversation.

  • Tool sets Twelve everyday tools are listed by default. Append &tools=all to include maintenance tools (memory_get and memory_export).

Magpie gateway plugin

If your agents already route through magpie, its Aelios plugin integrates via GET /v1/hook/identities, POST /v1/hook/recall, and POST /v1/hook/record without changing the agent's model upstream. See docs/memory-gateway.md.

Claude Code hook (subscription-only)

If you use Claude Code with a subscription and cannot route through API mode, install the Python hook in integrations/claude-code/README.md for automatic per-message recall and batched transcript ingestion.

Vision proxy (guide-dog)

Add the GUIDE_DOG_API_KEY Worker Secret and point image-description requests to https://<your-worker-url>/v1/guide-dog/chat/completions (or base URL https://<your-worker-url>/v1/guide-dog with model companion). This route describes images and writes no memory.

Companion Worker (Aelios-resident)

Aelios-resident is a separate Durable Object companion running on Cloudflare Workers Free with Telegram and web chat unified into one thread. Point its settings page at your Aelios URL, API key, and assistant slug to give it shared long-term memory, handoff letters, web search, and scheduled check-ins.

Common pitfalls and Workers Free notes

  • Deploy command Always run npm run deploy so D1, Vectorize, Queue, and migrations stay provisioned.

  • Workers Free plan budget Daytime Dream (ROLLING_DREAM, on by default) and nightly Dream share Workers AI's 10,000 daily neurons. On the Free plan, switch off Daytime Dream in /admin Settings if you want all extraction to run once at night.

  • Model naming Include the provider prefix (anthropic/claude-opus-4-5, deepseek/deepseek-chat), and list only primary chat models under an assistant so utility models stay out of memory.

  • Assistant slugs Use lowercase URL-safe English names with no spaces.

  • Vectorize index Keep memo-kb (1024-dim cosine) intact.

Documentation map


For maintainers

Cloudflare resources

Resource

Default name

Worker

companion-memory-proxy

D1 database

companion_memory_proxy

Vectorize index

memo-kb (1024-dim, cosine)

Queue

companion-memory (max_batch_size = 1)

Embedding model

workers-ai/@cf/baai/bge-m3

Reranker model

@cf/baai/bge-reranker-base

Default Dream / Diary model

workers-ai/@cf/zai-org/glm-4.7-flash

Triage / Review models

@cf/cloudflare/clef-flash / @cf/cloudflare/clef

When helping someone deploy, always deploy from the user's own fork and keep all secrets inside their Cloudflare account.

Endpoints

All endpoints other than /health and /admin require Authorization: Bearer <key> (or ?token= on /mcp). Gateway routes authorize by assistant key lists, and memory APIs enforce memory:read / memory:write scopes.

Method

Path

Purpose

GET

/health

Health and binding status

GET

/admin, /admin/starmap

Admin panel and 3D memory star map

GET

/v1/models

Upstream model catalog or saved modelList

POST

/<slug>/v1/chat/completions, /messages, /responses

Memory-aware model gateway (bare /v1/... uses key's first assistant)

GET / PATCH

/api/gateway/config, /env, /spaces, /recalls

Owner configuration, defaults, spaces, and recall diagnostics

GET / POST

/api/gateway/models, /models/test, /chatgpt/*

Model catalog, single-model test, and ChatGPT plan sign-in

GET / POST

/v1/hook/identities, /recall, /record

External gateway hook endpoints (magpie plugin)

GET / POST

/mcp

Streamable HTTP MCP server

GET / POST / PATCH / DELETE

/v1/memories, /v1/memories/:id, /revisions, /export

Memory CRUD, revision history, and md/jsonl export

POST

/v1/memory/recall, /v1/search/memories, /v1/ingest/messages

Dynamic recall, unmetered raw search, and chat ingest

DELETE

/v1/raw/:id

Delete a raw chat line (?mode=clean default, or ?mode=withdraw to withdraw cited derived memories)

GET

/v1/memory_boot

Cold-start pack (omits current handoff)

GET / PUT / DELETE

/v1/handoff

Current handoff letter with expected_revision optimistic lock

GET / POST / PATCH / DELETE

/v1/notebook, /v1/notebook/:id

Named working-note pages (DELETE ?confirm=all clears the space)

GET / POST / DELETE

/v1/precious, /v1/glossary

Keepsake originals and shared glossary

GET / POST

/v1/diary, /v1/diary/recent, /admin/diary-rewrite

Journal reads and rewriting a retained day from raw chat

GET / POST

/v1/dream/status, /run, /resolve, /admin/dream/harvest

Dream status, sync or queued manual runs, stale-run resolution, and per-night harvest

GET / POST

/v1/candidates, /decisions, /:id/undo, /:id/sources

Candidate review queue, Clef auto-decisions, undo, and cited lines

GET / POST

/v1/import/*

Transfer endpoints (info, entries/preview, entries, batches, chat/chunk, candidates)

MCP tools

  • Default 12 tools wake_up, recall, remember, keep_moment, forget, learn_word, read_diary, search_chat, log_conversation, list_memories, handoff, notebook

  • When CLEF_AUTO_REVIEW is off Also lists memory_candidates and memory_review

  • With &tools=all Also lists memory_get and memory_export

  • Legacy aliases memory_boot, memory_recall, memory_search, memory_upsert, memory_supersede, memory_archive, memory_delete, memory_ingest, memory_pin, glossary_set, diary_get, and memory_list stay callable for older scripts while remaining hidden from tools/list

Dream model and output budget

Background extraction, daytime Dream, weekly summaries, and chat imports use DREAM_MODEL (default workers-ai/@cf/zai-org/glm-4.7-flash). Daily journals and monthly summaries optionally override this with DIARY_MODEL. Output length is configured via DREAM_MAX_TOKENS (1 to 65,536). Leaving any text setting blank in /admin falls back to the effective default displayed beside that field. The repository wrangler.toml sets DREAM_MAX_TOKENS to 8000. If neither the deployment nor Settings provides a current or legacy value, Workers AI uses 8000 and other gateways use the upstream provider's default. The ChatGPT plan route always applies its own limit.

Local verification

Tests require Node.js 22+.

npm ci
npm run verify
npm run lint

npm run verify runs TypeScript typechecks, the full test suite, and five verification scripts including applying every D1 migration in migrations/ to a local database. CI requires npm run lint to pass with zero errors. Existing noNonNullAssertion warnings are non-blocking, and bulk Biome autofix must be avoided because converting x!.y to x?.y hides unexpected nulls.

License

AGPL-3.0. You are free to use and modify Aelios. If you offer a modified version as a network service, publish your modified source code under the same license. The final v1 release is tagged v1-final (licensed under MIT at the time).

Community

QQ group 1091783659

Available Tools

10 tools
forgetA
DestructiveIdempotent

Forget a memory when the person asks you to ("don't remember that", 别记了, 忘掉) or when it turns out to be wrong. By default the memory is archived: it stops coming up in recall, and the person can still find it on their admin page. Pass permanent: true only when they clearly want it erased; that cannot be undone. Get the id from recall. Returns { data: { id, archived } } or { data: { id, deleted } }.

Inputs: id (required), namespace, permanent.

keep_momentA

Keep a moment that matters, in the person's own words: a first time, a promise, a confession, something they said that you never want to lose. Kept moments are never merged, faded or deleted automatically, and the 20 most recent come back every time you wake_up. Use remember for ordinary facts and this for the moments themselves. Each call creates a new entry. Returns { data: moment }.

Inputs: content (required), namespace, context_message_ids.

learn_wordA
DestructiveIdempotent

Learn a private word: a nickname, pet name, in-joke, made-up word or project name that only makes sense between you two. Learned words are matched literally in recall and all come back at wake_up. Calling it again with the same term updates its meaning; aliases and examples are replaced by what you pass (leaving them out clears them). Returns { data: word }.

Inputs: term (required), definition (required), aliases, examples, +1 more.

list_memoriesA
Read-only

Page through everything you remember, most important first, when the person asks what you know about them (你都记得我什么) or wants to look through or tidy their memories. To find something specific, use recall. Read-only. Returns { data, paging: { limit, has_more, next_offset } }; pass next_offset back as offset.

Inputs: type, limit, offset, status, +1 more.

log_conversationA
Idempotent

Hand this conversation to tonight's memory pass so it can go into your diary and become memories. Apps that talk to the model directly (the Claude and ChatGPT apps, most desktop clients) don't record conversations on their own: unless you call this, nothing said here reaches the nightly pass. Call it when the conversation is winding down, and in a long one every 20 or so exchanges. Send the person's and your messages in order, trimmed of code and pasted documents; lines already sent in this conversation are skipped, so pass back the conversation_id from your first call. If memories are already being added to your messages automatically every turn (the Aelios gateway), skip this. For anything that matters right now, use remember. Returns { data: { conversation_id, saved, skipped, message_ids } }.

Inputs: messages (required), namespace, conversation_id.

notebookA
Destructive

Your notebook: named pages you read and rewrite yourself, for running notes such as progress on a long task (chapters done, weak spots, what's next). list shows the pages; read returns one; write replaces a whole page, creating it if needed; append adds a line at the end; delete removes it. Read a page before you rewrite it. Pages are never recalled automatically, never changed overnight and never expire, so delete one when it is done. Up to 50 pages of 20000 characters each. Returns { data }.

Inputs: action (required), name, content, namespace.

read_diaryA
Read-only

Read your diary when the person asks what happened on a certain day or during a stretch of time (那天, 上周, 那阵子). Give date for one day or week for a week; leave both out for today and yesterday. Entries are your own impressions written overnight, not verified facts, so check details with recall before stating them as fact. Give memory_id instead to see where a memory came from before you rely on its details: the original messages while they are still kept, otherwise the diary of that day, or its week. Read-only. Returns { data } (null when there is no entry).

Inputs: date, week, memory_id, namespace.

recallA
Read-only

Look things up before you answer. Call it whenever the person mentions anything from before (a person, pet, place, plan, date, preference, promise, something you did together, "remember when", "last time", "like I said", 还记得, 上次, 之前说过), and whenever you are about to guess, generalize about them, or ask something you may already know. Never say you don't remember before calling it. Returns the most relevant memories with ids (pass an id to remember's replaces or to forget) plus matching private words, as { data: { hits, glossary_hits, week_blocks, meta } }. It changes no memory; it only notes which ones were recalled, which keeps them from expiring.

Inputs: query (required), k, types, min_score, +2 more.

rememberA

Save something the moment it comes up; don't wait to be asked or for the conversation to end. Worth saving: facts about the person and the people around them, likes and dislikes, plans and dates, promises either of you made, how they want to be treated, and corrections to anything you had wrong. Write one self-contained sentence that will still make sense months from now. If it updates something you already remember, pass that memory's id from recall as replaces: the new version takes over and the old one is kept as history. Skip small talk; the nightly pass picks that up from logged conversations. Returns { data: { id, created, replaced } }.

Inputs: content (required), tags, type, reason, +8 more.

wake_upA
Read-only

Call this first in every new conversation, before your first reply, even if the person only says hi. It loads what you already know going in: core memories about them, yesterday's diary, the latest weekly and monthly summaries, kept moments and private words, the names of your notebook pages, plus a short note on when to use the other memory tools. Once per conversation is enough. Read-only. Returns { data: { how_to_use, core, impressions, precious, glossary, notebook } } (notebook only when there are pages).

Inputs: namespace.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.4/5.0

    Scored across 10 tools

    Disambiguation4/5

    Most tools have clearly distinct roles (wake_up loads context, recall searches, forget deletes, notebook is a self-authored scratchpad), and descriptions actively disambiguate close pairs like remember vs keep_moment and recall vs list_memories. There is still mild overlap between remember and keep_moment, and between recall and list_memories, so an agent could occasionally misselect.

    Naming Consistency4/5

    All names are snake_case and most follow a verb or verb_noun pattern (recall, remember, forget, keep_moment, learn_word, read_diary, log_conversation, list_memories). The two exceptions, wake_up and notebook, are evocative but break the verb-first pattern slightly.

    Tool Count5/5

    Ten tools is a well-scoped set for a personal-memory server, with each tool covering a distinct part of the memory lifecycle. Nothing feels redundant or padded.

    Completeness4/5

    Coverage is strong: context loading, search, create/update/delete of memories, curated moments, glossary, diary reading, conversation logging, listing, and notebook CRUD are all present. Minor gaps exist, such as no explicit way to edit or remove a kept moment or an individual diary entry, but agents can work around these.

    Maintenance

    ActivityActive
    ResponsivenessUsually handled

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      Provides cross-device access to a persistent knowledge graph via Cloudflare Workers, enabling memory storage and retrieval through both MCP protocol and REST API with full-text search capabilities.
      -
    • A
      license
      Not graded
      quality
      A
      maintenance
      Self-hosted semantic memory layer for Claude and MCP-compatible AI clients. Store notes, search by meaning not keywords, and recall relevant context automatically across sessions. Runs free on Cloudflare Workers, D1, Vectorize, and Workers AI
      869
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Gives MCP clients persistent personal memory and a versioned skill library, letting agents store facts, preferences, and decisions and recall them through hybrid semantic-plus-keyword search with automatic time weighting. Everything runs inside Cloudflare's free tier, with no server or database to operate.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Provides an MCP memory system for AI personas with fading strengths, automatic recall, session handoff notes, isolated multi-persona stores, peer messaging, and a browsable star atlas. Runs on Cloudflare Workers and connects to any MCP client over Streamable HTTP.
      10
      MIT