What Is a Genome?

A genome is a cognitive blueprint. When you spawn a creature, its genome is copied into the container. It defines the structure of the creature’s mind (what tabs appear, what tools exist, how validation works) but not its behavior. Two creatures with the same genome and different purposes will act completely differently.

Genomes are self-contained directories. They include source code, a genome.json manifest, and everything the creature needs to boot. No orchestrator-specific dependencies.

genome.json Reference

Every genome has a genome.json at its root:

Tabs

Each tab is an object:

Built-in Genomes

OpenSeed ships with three genomes bundled in genomes/. They’re the source of truth and available out of the box.

dreamer (default)

Full cognitive architecture. Creatures with this genome can dream, consolidate memories, self-evaluate, manage fatigue, browse the web with a persistent browser session, and maintain evolving rules about their world.

Tabs: purpose, diary, observations, rules, dreams, self-eval

Source: genomes/dreamer/

Use this when you want a creature that learns and adapts over time.

minimal

Bare-bones loop: bash and sleep. No memory, no dreams, no self-evaluation. The creature discovers everything on its own from a blank slate.

Tabs: purpose

Source: genomes/minimal/

Use this when you want to see what emerges without any cognitive scaffolding, or as a starting point for a custom genome.

wonders (experimental)

Subconscious memory experiment. No explicit memory, no observations, no rules, no dreaming. Conversation resets every cycle. The only source of long-term continuity is a background process — the subconscious — that watches what the creature is doing, hypothesizes about what past experience might be relevant, searches for it, and injects it as a thought if something genuinely useful turns up.

Tabs: purpose

Source: genomes/wonders/

Use this when you want to test implicit memory retrieval in isolation, or study what happens when a creature’s only connection to its past is associative recall. See Subconscious Memory for the full architecture.

Managing Genomes

seed genome list                                           # show installed + bundled genomes
seed genome search trading                                 # search community genomes on GitHub
seed genome install dreamer                                # install from openseed-dev/genome-dreamer
seed genome install someuser/genome-foo                    # install from any GitHub repo
seed genome install someuser/monorepo/genomes/foo          # subdirectory within a repo
seed genome install https://github.com/someuser/cool-mind  # full URL works too
seed genome remove foo                                     # remove an installed genome

When you spawn a creature with a genome that isn’t installed locally, OpenSeed will auto-install it from GitHub.

Creating Your Own

The easiest way to build a custom genome is to fork an existing one:

1. Start from genomes/dreamer or genomes/minimal in the openseed repo
2. Edit genome.json: change the name, tabs, validation command
3. Modify the source code to add or remove cognitive features
4. Push to your own GitHub repo
5. Add the openseed-genome topic to your repo on GitHub so others can discover it via seed genome search
6. Install it: seed genome install your-username/genome-your-name
7. Spawn a creature with it: seed spawn scout --genome your-name

Genomes have no dependency on the orchestrator. Everything the creature needs lives inside the genome directory. A genome could be written in any language as long as it exposes the expected HTTP endpoints (/healthz, /tick, /wake, /message).

If your genome requires a minimum version of OpenSeed, add a requires field to genome.json:

{
  "requires": { "openseed": ">=0.1.0" }
}

Users will see a warning if their OpenSeed version is too old.

Extracting a Genome from a Creature

Creatures evolve through self-modification. You can extract a creature’s evolved code back into a new genome:

seed genome extract eve --name evolved-minimal

This captures the creature’s self-modified source code, strips its identity and cognitive state, and packages it as a new genome with lineage tracking. The extracted genome.json includes a lineage field recording which genome it descended from, which creature it was extracted from, and at what point in the creature’s history.

The result is a genome that embodies everything the creature learned about how to think, without carrying over what it was thinking about. Anyone can spawn from it.