From standalone project to Flashforce infrastructure

Fishbrain taught AI how to remember. Then it became part of the workforce.

Fishbrain began as an independent attempt to solve a problem that stood between capable models and persistent AI employees: durable, scoped memory across conversations, projects, and time.

Its standalone development has concluded. The core ideas and architecture now live inside Flashforce.

Standalone product concluded in 2026 Architecture incorporated into Flashforce
The Fishbrain mascot: a smiling goldfish with an exposed pink brain, giving a fins-up pose atop a glowing globe of binary digits.
A cartoon fish whose translucent skull reveals turning gears and glowing lightbulbs.

Why Fishbrain existed

Before AI could become a coworker, it had to stop forgetting the job.

Frontier models were already capable. They were also stateless. Every new session started from zero, and the burden of continuity fell on the human in the chair.

A longer context window was not durable memory — it was a bigger short-term buffer that still emptied at the end of the conversation. A pile of chat logs was not governed organizational knowledge. It was a transcript nobody could trust, correct, or scope.

Fishbrain was built to answer five questions:

  • What should become memory?
  • Where should that memory live?
  • Who or what should be allowed to use it?
  • How should corrections replace outdated information?
  • How can the same persistent context survive a model change?

What we built

More than a vector database with a fish on it.

Six pieces did the real work. Each one existed because storing embeddings alone never produced memory anyone could rely on.

A cartoon fish with visible brain gears, surrounded by bubbles that each contain a smaller reflected fish.
  1. 01

    Scoped memory

    Global, domain, and topic boundaries kept unrelated context from bleeding together.

  2. 02

    Atomic factlets

    Conversations and source material became compact, independently useful pieces of knowledge rather than transcript sludge.

  3. 03

    Reflection

    New conversations could propose durable memories, corrections, and higher-level insights.

  4. 04

    Retrieval and context injection

    Relevant memories were selected, ranked, and inserted into model context only when useful.

  5. 05

    Provenance and correction

    Memory carried source, trust, history, and supersession rather than silently overwriting the past.

  6. 06

    Model independence

    The memory layer belonged outside any one model provider.

The architecture

A memory system, not a longer prompt.

Fishbrain treated memory as a governed lifecycle: capture, structure, retrieve, correct, and reuse.

  1. Input Conversation, file, or correction
  2. Extract Candidate factlets
  3. Govern Scope, trust, and provenance
  4. Persist Durable memory
  5. Select Relevant retrieval
  6. Inject Model context
  7. Result Better continuing work

What happened next

Fishbrain solved the prerequisite.
Flashforce became the product.

Three gold silhouettes on dark green: a small fish, then a larger fish, then an upright walking fish-person — the project growing into something that can do a job.

Fishbrain asked

How can AI remember?

Flashforce asks

How can an AI employee hold a job, continue work, use tools, respect authority, produce artifacts, and preserve a trustworthy record?

As Flashforce evolved, memory became one internal substrate among identity, responsibilities, tools, work state, authority, approvals, artifacts, and audit truth.

Maintaining Fishbrain as a separate application and database no longer improved that mission. Its architecture belonged inside the system it had helped make possible.

Fishbrain's legacy inside Flashforce

Memory is now built into the employee.

Flashforce builds AI employees for businesses. Fishbrain's durable-memory architecture now works inside those employees alongside identity, role, company context, tools, authority, approvals, work continuity, artifacts, and audit truth.

Memory is no longer a separate destination. It is part of how a Flashforce employee understands the job, continues the work, and remains accountable over time.

  • Persistent employee identity
  • Role and responsibility
  • Company and relationship context
  • Work continuity
  • Tools
  • Authority boundaries
  • Approvals
  • Inspectable artifacts
  • Audit truth

Timeline

Three chapters.

  1. 2025

    Fishbrain began as an independent scoped-memory project.

  2. 2025–2026

    The architecture expanded through semantic retrieval, reflection, context engineering, correction, and organizational scoping.

  3. 2026

    Standalone development concluded and the architectural lineage moved into Flashforce.

A cheerful blue cartoon fish with a big brain, smiling.

The fish did its job.

Fishbrain was built to give stateless AI continuity. That work now continues inside a larger system designed not merely to remember conversations, but to carry responsibility over time.

We didn’t build the brains.
We built the hippocampus.

Meet Flashforce