# Loominary: a local archive for chats that currently live in a scrollbar

> A browser-local conversation manager for seven AI platforms whose real contribution is not the search box but the branch tree: it records every regeneration and prompt edit in the background, so a version the platform already overwrote is still somewhere you can find it.

**Laumss/loominary** — Conversations shouldn't just live in a scroll bar. Loominary gives your Claude and SillyTavern conversations a home — a local archive of every conversation, yours to keep. Coming soon... 

- Repository: https://github.com/Laumss/loominary
- Website: https://laumss.github.io/loominary/welcome
- Stars: 460 · Forks: 49
- Language: JavaScript
- License: MIT
- Published: 2026-09-14 · Updated: 2026-09-14 · Language: en
- Canonical page: https://hysenlabs.com/projects/laumss-loominary

## The problem is stated in the first person

The motivation paragraph is unusually direct. Hundreds of chats were scattered across Claude, ChatGPT and Gemini with no good way to search any of them, which is why the tool exists. Everything else follows from that: Loominary is local-first, runs entirely in the browser, and keeps data on your machine. The privacy section is categorical rather than promissory, with no analytics, no telemetry and no data collection, and the extension only activates on supported platforms and communicates solely with your local Loominary instance. That last clause matters more than the rest, because an archive tool that phoned home would defeat the purpose. The online version at laumss.github.io is a browser app, and a Greasyfork userscript is listed for people who want it injected into the chat page itself.

## Seven platforms, three columns of honesty

The support table is worth reading as a grid rather than a list, because the columns are load chats, project and memories, and realtime branches. Claude is the only row with a check in all three. ChatGPT, Gemini and Grok load chats and support realtime branches but carry no project context. NotebookLM loads chats and its branch column reads Whiteboard mode rather than yes or no, which is a different mechanism wearing the same column. Google AI Studio has a tick for loading and nothing in the other two columns. SillyTavern has checks for loading and for project context, and its branch column reads Merged branch files. One caveat sits next to the table: the userscript version does not support SillyTavern, so you load files and directories manually or use the Chrome, Firefox or Edge extension.

## Realtime mode exists because saving destroys the old answer

Every platform overwrites the previous response the moment you edit a prompt, so once you hit save the old answer is gone along with the comparison you wanted to make. Realtime mode records every version in the background instead: every regeneration, every prompt edit. Opening the preview shows the branches laid out on a tree, side by side. The reason this matters is described concretely, with platform updates leaving people unable to revert when reworking a prompt produced something worse, and with image generation where comparing four variations means keeping four answers the platform would otherwise discard. The search side closes the loop, since the global index covers every branch, which means content from an old revision stays findable even after the platform itself has buried it. Tags on messages persist across sessions and survive into exports.

## Merging SillyTavern branches is the part SillyTavern lacks

The SillyTavern support goes beyond reading chats. Loominary archives character cards, world books and the current preset alongside the conversations, and it loads all branch JSONL files and merges them into a single timeline, which the README describes as a feature SillyTavern does not natively support. That is the difference between an importer and an archivist: the branches are the interesting part of a roleplay session and the platform keeps them in separate files. ChatGPT gets the memory export beside the conversations. Claude gets the widest capture of the three, taking in project descriptions, instructions, project memory, knowledge files, user preferences and saved memories alongside the chat itself. Search then runs across all of that by content or title, or semantically if you connect your own embedding model, and filters can narrow to images, thinking processes, Artifacts or tool calls.

## Exports: one format live, two struck through

The export section is where the README admits something. Markdown export is live. PDF with LaTeX and image support is struck through, and so is long screenshots, both with the strikethrough left in place rather than removed. What you can include in an export is a choice rather than a default: timestamps, thinking processes, Artifacts, tool calls or citations, which matters if your archive contains reasoning you would rather not hand over. Batch processing handles scale, exporting hundreds of conversations into a ZIP, or only the latest branch of each when the full history is unnecessary, and the README names massive Claude account exports as the case that makes this matter. Claude's native conversation favorites are preserved, and you can add your own on top. Branch navigation is automatic, with the latest branch one click away.

## The manifest still remembers the old name

package.json tells you what the rewrite did and did not clean up. The package is Loominary at version 26.03.0, a calendar-style version rather than a semver one, marked private, and its description is still A handy tool for exporting chat conversations. The start script sets PUBLIC_URL to /lyra-exporter, the project's former name. The deploy:pages script publishes to a repository called react, which is not this one. The release tags are v7.5, v7.4 and v7.2 while the packages inside them are numbered 1.7.5, 1.7.4 and 1.7.3, so the tag and the version inside it disagree in both prefix and number. There are tauri scripts with no Tauri configuration in the tree, and a build.py sitting beside the npm scripts. The dependency list maps cleanly onto the feature list: @xenova/transformers for the optional local embeddings, katex and jspdf for the PDF path, jszip and fflate for archives, html2canvas for screenshots.

## Announced, not shipped

Several things the project describes do not exist yet, and the page is upfront about each. Mobile support is in the works, with navigation that hides on scroll, a dedicated mobile search layout and back gestures. The developer path, where a local build adds a backend that serves archived data to local AI clients over MCP, is described as still taking shape, with the architecture supporting it but the integration surface in its early stages. The contributing guide and development roadmap are on the way, with discussions invited in the meantime. The repository description itself ends with Coming soon. Against that, the last push to main is dated 25 March 2026 and the newest release, 1.7.5, was tagged on 2 March 2026, so this is a project with a shipped feature set that has been quiet for a few months rather than one abandoned mid-build. The claim that no other tool covers all four platforms for full account export is the author's own, offered as a personal assessment rather than a survey.

## Conclusion

Loominary fits someone with a few hundred AI conversations scattered across three or four services and no way to search them, since the archive runs entirely in the browser and stores nothing remotely. It does not fit someone who needs PDF or long-image exports today, because both are struck through on the visible page and only Markdown is offered, and it does not fit someone expecting a maintained release stream, with the last push to main dated 25 March 2026 and the newest tag from 2 March. Before you rely on it for SillyTavern, note the userscript build does not cover that platform and you need the Chrome, Firefox or Edge extension instead. And if you intend to expose the archive to a local agent over MCP, that integration is described as early stage, so treat the architecture as ready and the surface as not.

## FAQ

### What does Loominary do with my AI conversations?

It loads them from Claude, ChatGPT, Gemini, Grok, NotebookLM, Google AI Studio or SillyTavern into a local archive you can search, tag, walk through branches and export. Everything runs in the browser, and the project states there is no analytics, no telemetry and no data collection.

### Does Loominary keep old versions of a response?

Realtime mode does. It records every regeneration and every prompt edit in the background, and the preview lays the branches out on a tree side by side, which is the workaround for platforms that overwrite the previous answer the moment you edit a prompt and save.

### Which platforms does Loominary pull project context from?

Claude, where it archives project descriptions, instructions, project memory, knowledge files, user preferences and saved memories; ChatGPT, where it exports your ChatGPT memories; and SillyTavern, where it archives character cards, world books and the current preset and merges all branch JSONL files into a single timeline.

### Can Loominary export to PDF or a long screenshot?

Not according to the visible page. Markdown export is live, while PDF with LaTeX and image support and long screenshots are both struck through. What you can choose to include is timestamps, thinking processes, Artifacts, tool calls or citations.

### Is Loominary still being worked on?

The last push to the main branch is dated 25 March 2026 and the newest release, 1.7.5, was tagged on 2 March 2026. Mobile support is described as in the works, the MCP backend for local AI clients as an early-stage integration surface, and the contributing guide and roadmap as on the way.

## Sources

- [Laumss/loominary on GitHub](https://github.com/Laumss/loominary)
- [License: MIT](https://github.com/Laumss/loominary/blob/main/LICENSE)
- [Project website](https://laumss.github.io/loominary/welcome)
- [README](https://github.com/Laumss/loominary/blob/main/README.md)
- [Releases](https://github.com/Laumss/loominary/releases)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/laumss-loominary