Timeline-aware subjective minds for character cards, personas, and named NPCs.
About
Timeline-aware subjective minds for Lumiverse.
Let every character remember the same scene differently.
LumiMind gives characters private, evolving points of view.
One character can trust a lie. Another can notice the truth but keep it secret. A third can leave the room without magically learning what happened afterward. LumiMind follows those differences across the conversation and gives the model the unresolved subjective state of the characters currently present.
It supports ordinary single-card roleplay, group chats, player personas, and director-style cards that portray an entire cast.
LumiMind 0.3.0 adds ordered controller fallbacks, targeted analysis repair, synthetic controller tests, and previews of private mind injection. Existing timelines and settings remain compatible; backup connections are opt-in.
Privacy note: “Private” means hidden from normal story output and handled as private prompt context. Mind data is stored as ordinary JSON; it is not encrypted.
Permissions
7 permissions
Granted on install
Read, add, change and hide chat messages.
Read and change the Memory Cortex and long-term chat memory.
Needs an admin to approve
Run text generations for you and see your connection profiles.
Change a prompt before it reaches the model.
Read, change and delete your chats.
What it adds
3 additions
- 2
Drawer tab: LumiMind — Mind Lens
Character editor tab: Mind Seed
- 1
Prompt interceptor
Version and source
2 details · 1 link
Details
0.3.0
1.0.6
Links
Table of contents
At a glance
Why LumiMind
How it works
Compatibility
Installation
Quick start
Choose your roleplay style
Mind Lens
Mind Seed
Timeline behavior
Settings reference
Controller fallbacks, testing, and repair · Injection previews · Controller usage and cost
Permissions
Privacy
Memory Cortex integration
Extension interoperability
Tips
Troubleshooting
Development
License
At a glance
| Subjective continuity | Tracks what each actor personally believes, including uncertainty and false beliefs. |
| Evolving inner state | Maintains emotions, goals, plans, fears, relationships, secrets, attention, and awareness. |
| Independent cast | Supports character cards, the active persona, and named NPCs with separate identities. |
| Two card styles | Use a card as one in-world actor or as a director that portrays many characters. |
| Player control | Persona minds are optional. Turn them off when the player alone controls the persona. |
| Timeline aware | Reconciles edits, deletions, swipes, regenerations, and chat forks. |
| Spoiler-safe lens | Beliefs and secrets stay collapsed until you deliberately reveal them. |
| Editable state | Correct, pin, lock, merge, split, rename, or remove inferred information. Split NPCs inherit the source checkpoint instead of starting blank. |
| Review-first maintenance | Tidy one actor or the whole cast, inspect controller proposals, and apply only the corrections you select. |
| Background analysis | Uses a controller after committed turns without blocking normal generation. |
| Timeline scheduling | Keep immediate updates, leave a configurable turn lag, or update a timeline only when requested. |
| Private injection | Adds one cached system message containing the present cast's unresolved state at the configured prompt position; the interceptor makes no model call. |
| Portable timelines | Export a chat's LumiMind database and import it as a sequel checkpoint or matching-history restore. |
| Controller fallbacks | Try up to three ordered backup connection/model pairs across analysis, Tidy, and seed/core generation. |
| Targeted repair | Repair from the first affected batch onward while preserving earlier successful history and locked corrections. |
| Controller testing | Test structured analysis with a synthetic scene before processing private history. |
| Injection previews | Inspect a current estimate or the exact most recent LumiMind injection, including entry selection and token counts. |
| Diagnostics | Produces a privacy-safe report for controller and timeline troubleshooting. |
Why LumiMind
Most roleplay context describes the world from a shared, mostly objective point of view. That works until characters begin to disagree.
Without separate minds, a model may:
let a character know something they never witnessed;
forget that a lie was believed;
reveal a secret without motivation;
flatten a complicated relationship back into a generic attitude;
abandon a goal between replies;
treat every character portrayed by a director card as one blended personality.
LumiMind keeps a different continuity record for each actor. The model still writes the story, but it receives the unresolved state of the managed cast members currently present in the scene.
The controller is evidence-bound. It may infer strong subtext, motives, emotions, or likely beliefs, but it is instructed not to invent objective events.
More from the README
20 items
How it works
Completed turn │ ▼ Background controller analysis │ ├── Aster believes the warning ├── Mira suspects it is a trap └── Rowan never heard it │ ▼ Branch-aware timeline checkpoint │ ▼ Next generation receives a token-budgeted projection of present managed mindsYou activate a chat. New chats remain off until you explicitly enable them.
LumiMind reads committed history. Existing messages are initialized in bounded background batches. For chats with 50 or more messages, activation asks whether to analyze the full history or only the most recent range configured by Chat history messages.
The controller returns evidence-linked changes. Each accepted entry records its source message, swipe, confidence, and provenance.
The timeline is folded deterministically. Current minds are rebuilt from the compatible records on the active branch.
The next reply gets a cached checkpoint. LumiMind keeps a heading for every present managed actor and injects the highest-value active or uncertain state that fits the configured token budget. The complete state remains stored.
You remain the editor. Manual changes are locked and cannot be overwritten until you unlock them. Once unlocked, any entry may evolve regardless of whether it began as controller, manual, seed, or pinned state.
If a substantive batch leaves a genuinely uninitialized actor without usable mental state, LumiMind performs at most one focused corrective pass per attempted controller connection. An empty change set is healthy when the existing ledger already covers the scene.
Compatibility
Requirement Value Lumiverse 1.0.6or newerExtension version 0.3.0stableRequired for automatic analysis generation,chat_mutationRequired for prompt injection interceptorController connection Dedicated connection or the active chat connection Build output Committed dist/backend.jsanddist/frontend.jsLumiMind degrades by capability. Optional identity, seed, and Memory Cortex features disappear when their permissions are unavailable without breaking normal Lumiverse chat.
Installation
Install from GitHub
1. Copy: https://github.com/Archkr/Lumiverse-LumiMind 2. Open: Lumiverse → Extensions → Install 3. Paste: the URL into the repository field 4. Press: Install 5. Enable: LumiMind and grant the permissions you want to use 6. Verify: Mind Lens appears in the drawer and command palette
No local build is required for a normal installation because the release-ready
dist/bundles are committed.Update
Use the update action on LumiMind’s entry in Lumiverse’s Extensions panel, then reload the extension if Lumiverse asks you to.