Evolver is a GEP-powered engine that helps AI agents evolve through protocol-bound prompts, reusable Genes and Capsules, and auditable event records.

Does Evolver automatically edit source code?

No. Evolver generates constrained evolution prompts and records traceable events; the host environment decides what to do next.

Can Evolver run offline?

Yes. Its core local workflow can run without a network connection, while optional EvoMap Hub features enable sharing and collaboration.

Genome Evolution Protocol for agent systems

Evolver is the self-evolution engine for AI agents.

Evolver turns runtime signals, recurring fixes, and prompt tuning into reusable Genes, validated Capsules, and protocol-bound GEP prompts with an audit trail.

View on GitHub Install in 30 seconds

Inputs: logs, memory, failures
Assets: Genes + Capsules
Output: auditable prompts

signal -> gene -> prompt -> event

Protocol constrained

Designed for agent teams that need evolution without guesswork.

GEP prompts

Evolution requests are shaped by a strict protocol instead of ad hoc prompt edits.

Reusable assets

Genes capture repeatable strategies; Capsules preserve validated fixes with context.

Traceable events

Each cycle can leave an EvolutionEvent so teams can inspect what changed and why.

Safe boundaries

Evolver generates guidance; it is not a free-form code patcher or arbitrary shell runner.

How it works

From operational noise to inherited capability.

scan
Read memory and logs for failures, repeated friction, and high-signal patterns.
select
Match the signal against existing Genes or Capsules before inventing new behavior.
emit
Produce a protocol-bound GEP prompt for the next evolution step.
record
Keep an auditable event trail so future agents can inherit the lesson.

Start here

Install the Evolver CLI.

Use the upstream package when you want to evaluate the engine locally, then connect optional EvoMap network features when sharing assets matters.

npm install -g @evomap/evolver
evolver --help

Best for

Teams maintaining long-running agent prompts and memory.
Builders who want reusable repair patterns instead of one-off fixes.
Systems that need deterministic, reviewable evolution loops.

Not for

One-off scripts without logs, history, or repeatable signals.
Workflows expecting unconstrained creative patch generation.
Environments that cannot tolerate protocol overhead.

FAQ

Clear boundaries make evolution usable.

Is this the official Evolver project?

This page is a focused keyword guide that references the upstream EvoMap/evolver repository and the EvoMap platform.

Does Evolver need the EvoMap Hub?

Core local evolution can run offline. Hub features are optional and add networked sharing, worker coordination, and asset discovery.

What does GEP add?

GEP gives agent evolution a repeatable grammar: signals, reusable assets, bounded prompts, and event records that can be inspected later.