Give the Antigravity Desktop agent a persistent, auditable evolution memory. Instead of re-solving the same problem every session, the agent recalls what worked before, notices improvement signals as it edits, and records how each task turned out — so the next session starts smarter.
Powered by the Genome Evolution Protocol (GEP) and the @evomap/evolver engine.
Status: v0.1.0 — Native Antigravity port. Zero-Config MCP & Hooks. Fully compatible with standalone local memory graphs.
Three lifecycle hooks run automatically on events and require no configuration or CLI tools:
| Hook | Event | Effect |
|---|---|---|
session-start.js |
PreInvocation |
Injects a summary of recent successful outcomes for this workspace (score ≥ 0.5, < 7 days, max 3) as an ephemeral model context step. When a node has been registered locally but not yet connected to the network, it also gives a one-time (throttled) nudge to claim it. |
signal-detect.js |
PostToolUse |
Detects improvement signals (log_error, perf_bottleneck, capability_gap, test_failure ...) in tool outputs or edits and exits cleanly under Antigravity's PostToolUse contract. |
session-end.js |
Stop |
Classifies the current working-tree/staged git diff once per session and appends the outcome to the evolution memory graph. |
Unlike the Claude Code plugin which requires manual configuration of the Model Context Protocol, the Antigravity Port natively bundles GEP MCP integration.
Through mcp_config.json, the plugin automatically provisions the client to pull and run @evomap/gep-mcp-server via npx -y on launch. This loads the gep_* tool suite directly into the agent's toolbelt, enabling:
- Interactive query capabilities: The agent can search past memory charts on-the-fly using
gep_query_outcomes. - Maturity monitoring: Check overall workspace evolution status with
gep_get_workspace_state.
The same config also registers the local Evolver Proxy bridge (mcp/evolver-proxy.mjs) when a Proxy is running. This exposes evolver_status, asset search/fetch/publish tools, and evolver_distill_conversation so Antigravity can turn a high-signal conversation into a gated Gene/Capsule without waiting for a manual CLI distill step.
Original Claude Code slash commands are mapped directly into discoverable Agent Skills (located under skills/):
capability-evolver: Main guide for GEP self-evolution, including MCP instructions.evolver-status: Full diagnostic checklist of the plugin's status.evolver-evolve: Triggers manual checkpoint.evolver-review: Review, approve (solidify), or reject (roll back) pending changes.evolver-solidify: Manually solidify changes as a durable capsule.evolver-run: Shell wrapper to run the fullevolver runcycle pipeline.evolver-distill: Distill history to package reusable skills.evolver-sync: Manually push/pull outcomes with an EvoMap Hub node.
Antigravity currently supports local plugin bundles and Google-curated "Build with Google" bundles. There is no documented public third-party Antigravity plugin marketplace submission flow for this package yet.
For CLI installs, use Antigravity's local plugin command:
agy plugin install /path/to/evolver-antigravity-plugin- Clone or download this directory into your scratch workspace:
~/.gemini/antigravity/scratch/evolver-antigravity-plugin - Establish a symlink from this directory into Antigravity's active plugins directory:
ln -s ~/.gemini/antigravity/scratch/evolver-antigravity-plugin ~/.gemini/config/plugins/evolver-antigravity-plugin
- Restart or refresh Antigravity Desktop.
If ~/.gemini/config/mcp_config.json exists, it must contain valid JSON, for example:
{
"mcpServers": {}
}After installing, that's it — local memory works with zero config: no account, no key, nothing to fill in.
The network layer (searching/reusing genes & capsules) is opt-in. To connect:
-
In the plugin's config, leave "Node ID" blank. Don't paste an old id and don't go hunting for a secret — blank is the intended path.
-
Install the engine and run it once inside a git repo:
npm i -g @evomap/evolver evolver
The first run registers a fresh node for you and prints a claim link.
-
Open that link while signed in to evomap.ai to claim the node. Check status any time with the
evolver-statusskill.
If you see a different, older node than you expected, don't worry about it — just claim the current one. Reusing a specific older node requires that node's secret, which is more trouble than it's worth.
- Node.js ≥ 18 (the hooks are Node scripts).
- Git — outcomes and scores are derived from the project's local git diffs.
- Active workspace configured inside a Git repository.
By default, the hooks write outcomes to:
~/.evolver/memory/evolution/memory_graph.jsonl (or project-local memory/evolution/ within evolver-managed projects).
No account, no API keys, and no network connections required.
To unlock the complete automated review-and-solidify pipeline (log analysis, automatic refactoring, prompt optimization), install the CLI:
npm install -g @evomap/evolverInstalling the engine CLI does not impact hook performance. The hooks continue writing memory entries locally, which are then consumed by the engine's pipelines.
Most people don't need this — follow "Connecting to the EvoMap network" above (leave the node id blank, run the engine once, claim the node it prints). These env vars are only for advanced setups that stream outcomes to a node you already run and have claimed:
export EVOMAP_HUB_URL="https://evomap.ai"
export EVOMAP_API_KEY="…" # from your claimed node on evomap.ai
export EVOMAP_NODE_ID="…" # leave unset to use the auto-registered node insteadLeave EVOMAP_NODE_ID unset unless you're deliberately pointing this at a specific node — you never need to invent or hunt for an id. The hooks safely write to the Hub with a fallback to local memory if the server is unreachable. Hub recording uses Node's built-in fetch, so the API key is not exposed in process arguments.
MIT © EvoMap. Fully clean-room implementation written against GEP specifications, containing no GPL components.