OpenAlma

Gives an AI companion a soul. Local-first.

View the Project on GitHub

OpenAlma

SillyTavern Chat Client

How the soul behaves in SillyTavern — the controls, and letting her author herself. For installing the plugin and extension, see Getting started.

Controls in the extension

Control Location What it does
Memorize Now button memU extension panel Extracts the current conversation tail (everything after the last memorized point) without waiting for a sleep gap. Sends tail=true to the server. Disabled when no character is selected, memU is off for the card, the chat is a checkpoint or branch, or a memorize is already running.
Re-memorize chat SillyTavern’s chat options menu (the rotate-left icon) Developer testing only. Sends rebuild=true: archives the Soul’s whole database and starts a new one from this chat alone. Other SillyTavern chats and other clients’ memories are not included. Not a repair for bad extraction; use Atomic to correct memories.
Eye icon (👁) memU extension drawer header, next to the memU logo Opens a memory inspector. Each category shows as a memU lorebook holding the prose she’s written about it, not a list of raw entries.
Narrative Suggestion input memU panel, under the Memorize Now button Sends the soul a suggested revision of her narrative_self. See below.
Swipe / delete her last reply ST’s normal swipe or message delete Regenerating re-runs the turn — and the working state that reply wrote (her scratchpad and intentions at that moment) rolls back with it, one step.

Persona And AI Settings

Choose a persona on the saved character card before turning memU on. Each Soul has one fixed memU persona, including while memU is off. Existing Souls select their bound persona automatically. The light-blue memU switches on the card and in the panel control the same setting; the card’s memU-logo button opens the panel.

The panel reads mcp-memu-server/config.json, including hand edits, and can edit its model settings. Choose your configured OpenAI-compatible providers through ST; unsupported provider formats are unavailable. SillyTavern Replies (SRMM) stays in ST and controls ST replies, APImw synthesis and Narrative Suggestion. Other OpenAlma App Replies edits the shared server connection, except Iris’s voice model.

Advanced Mapping chooses models for memory work and copies those connections into the shared config, so they work with ST closed. Unchecking it asks for confirmation before removing those overrides, including any hand edits; ST retains profile/model choices for re-enabling. A memory step set to SRMM, ST selected or a named profile keeps following that source’s later changes while SillyTavern is open; config.json keeps the server’s own values. Use Claude Code switches server work to Claude Code while SRMM remains an independent API choice; hiding controls in Claude Code mode does not disable them. The Launcher explains context-window setup for Memorize.

Turning memU off restores plain ST replies, without deleting the Soul or changing their memU persona binding. ST’s native persona controls remain usable; no uninstall is needed.

Server Controls

Use Start and Stop in the memU Backend panel to control the shared memory server. Auto Start is off by default, including after upgrading from v0.1.22; turn it on again if wanted. It starts MCP with the SillyTavern service only when MCP is off, then requests graceful shutdown when ST stops. A server already running is left alone. Closing the browser tab does not stop ST.

With Auto Start off, start MCP in this panel or OpenAlma before chatting. Background requests never start it. Stop refuses an active Iris conversation; if Iris activity is unknown, manual Stop asks for confirmation and automatic shutdown leaves MCP running.

Memory section checkboxes

Toggle Default What
Override Summarizer on If on, replace SillyTavern’s summary message with memU’s. If off, memU’s renders alongside it.
Import Lorebooks on Publishes memU categories as SillyTavern lorebooks named memU - <Character> - <Category>, so the soul’s knowledge shows up in ST’s world info. Unchecking deletes any existing ones for this character.
Mental Health Addon off, per Soul Applies to this Soul’s SillyTavern chats, saved in this browser. When the conversation touches a relevant theme, she can draw on one of 15 curated coping entries (rumination, grief, panic, self-criticism, loneliness, etc.). Other surfaces such as WhatsApp follow the server’s retrieve.mental_health_query setting, off by default.

Relationships

The Memory section has a Relationships area (greyed out until a soul/character is active). Here you declare third parties the soul should be aware of — family, friends, coworkers, pets. Each entry becomes a named entity in the memory graph. When the soul extracts memories from conversation that mentions a declared relationship, she can attribute the memory to the right person rather than guessing.

You can add, edit, and soft-delete relationships. The section shows a warning when you exceed 20 entries.

Entities themselves are managed in the Atomic Mind Map, where each one has a permanent identity that survives renaming.

Letting the soul author her own self-model

The companion has a narrative_self, her evolving sense of who she is. Consolidation can revise it as her experience accumulates. You can also ask her to reconsider it through Narrative Suggestion, or correct it in Atomic.

Her character Description follows that shared self-description until you edit Description, Personality, Scenario or Example Dialogue. After an edit, the Description stays fixed. Clear Description to restore following; the other fields are left alone and still affect ST replies.

Your manual card affects her replies in ST only. Her memory work, autonomous activity and other apps still use her shared self-description. A handwritten card is not overwritten on first use.

Using Narrative Suggestion

  1. Open the memU extension panel.
  2. Type your suggestion in the Narrative Suggestion input — a phrasing, a correction, a new way of seeing herself.
  3. Click Send. A green check means she considered the request, not necessarily that she accepted it. A red ! means an error; mouseover explains it.
  4. If she accepts, her shared self-description changes and the previous version is preserved in memory. A following ST Description refreshes to match; a manual card stays untouched.
  5. Suggestions remain available with a manual card. Clearing Description restores syncing, not permission for her to consider a suggestion.

The panel waits 10 minutes between suggestions (shared across characters in the same browser), so the soul isn’t churning her identity every minute. For editing her memory directly, see Shaping her and the mind map.