Home / AI Labs / SeraLab

External memory & retrieval

SeraLab

Nothing worth remembering is lost.

Local research build

FRAGMENTSSTORED · LINKEDRETRIEVE CHATDOCUMENTNOTEMEMORYFILE query › SOURCE 1SOURCE 2SOURCE 3
Concept visualization

Overview

A memory engine that gathers scattered conversations, documents and notes, stores them with their sources, and links related pieces. When a question arrives, it retrieves only the fragments that matter — by keyword, by meaning, or both.

Why it exists

AI tools forget. Context is rebuilt by hand in every new session, and the sources behind a decision quietly disappear.

SeraLab is an external memory the user owns — kept locally, with originals and sources preserved, so any AI system can be handed the right context on purpose.

How it works

Material goes in as originals — conversations, documents, notes — and keeps its source. From that, SeraLab maintains working memory: decisions, open loops, handoffs between sessions.

When a task starts, it assembles a context or bootstrap pack: only the pieces that matter, each traceable back to where it came from.

Core capabilities

  • Originals & sources

    Raw material is preserved; nothing is summarised away without a trail.

  • Decisions & open loops

    What was decided, and what is still unresolved, are tracked explicitly.

  • Handoffs

    A session can leave a note for the next one — human or AI.

  • Context / bootstrap packs

    Selective recall builds a compact starting pack for a new task.

  • Evidence tracing

    Each recalled piece points back to its original source.

  • Backup

    Local data can be backed up and moved with the user.

Technical approach

  1. 01Originals in
  2. 02Store with source
  3. 03Index: full-text + semantic
  4. 04Hybrid retrieval
  5. 05Context pack with evidence
  6. 06AI session
Concept diagram — simplified, not a system screenshot.

  • Full-text

    Finds the exact words. Precise, but misses things said differently.

  • Semantic

    Finds related meaning, even with different words. Broad, but can drift.

  • Hybrid

    Combines both, then ranks — exact matches anchor, meaning widens the net.

Practical applications

  • Starting a new AI session without re-explaining the whole project.
  • Checking why a decision was made, with the source attached.
  • Handing work from one session — or one tool — to the next.

Development journey

  1. Began as a way to stop losing context between AI sessions.
  2. Added decisions, open loops and handoffs as first-class memory.
  3. v0.8: local operational build, with a portable release candidate reported by the developer.

Current status & limitations

Implemented

  • Local-first storage of originals with sources
  • Memory: decisions, open loops, handoffs
  • Full-text, semantic and hybrid retrieval
  • Context / bootstrap packs and backup
  • Offline mode — a model provider is optional

Planned / not yet active

  • Connection with Sereaniel — designed to integrate; currently partial / planned
  • Portable packaging beyond the release candidate

SeraLab is a personal research build, not a commercial product. Semantic retrieval finds related meaning, not guaranteed truth — evidence tracing exists so recalled context can be checked.

Future direction

  • Designed to serve Sereaniel as its memory layer.
  • Sharper selective recall: less context, more relevance.
  • Portable, user-owned memory that moves between tools.

The visual above is a concept visualization, not a product screenshot. No private data is shown.

← Back to AI LabsSeraYun home