Gemini Local Hub is a Next.js application that fronts @google/gemini-cli-core—the same engine as the Google Gemini CLI—and exposes it as a local HTTP API plus a chat UI.
Each plain gemini CLI invocation pays a cold-start cost (process boot, OAuth/session handling, model discovery). The Hub keeps warm GeminiClient instances in a process-wide registry keyed by trusted project directory, so follow-up requests from bash scripts, editors, or other tools avoid that boot tax. Pair this with POST /api/chat/start (or the examples/connect.sh helper) to warm a session before you fire prompts.
Authentication follows the Gemini CLI: sign in with gemini login (or your existing CLI OAuth). The Hub does not ship its own API keys; it reuses the CLI core against your account. It also normalizes multimodal input: multiple images are stitched server-side into a single composite when required by the model/API constraints, with prompts adjusted so vision is used directly.
In short: one long-lived local process, OAuth via the official CLI stack, HTTP in / multimodal out—optimized for developers who want repeatable, low-latency local automation.
| Capability | Summary |
|---|---|
| Stateful sessions | Conversation state and CLI session live across HTTP requests (singleton registry per trusted folder + optional sessionId). |
| Tool execution | YOLO mode runs tools on the server and continues the agentic loop; otherwise the stream pauses for approval (UI or POST /api/chat/tool). |
| Ephemeral turns | ephemeral: true clears history before a request—ideal for one-shot scripts (e.g. commit message generation) without polluting chat. |
| Streaming | Long-lived responses emit newline-delimited JSON events (Content-Type: application/x-ndjson) when streaming is enabled. |
| Trust model | Only trusted absolute folderPath values are accepted; manage trust via /api/registry/*. |
- Next.js (App Router) — API routes under
app/api/**/route.tsand the Hub UI. @google/gemini-cli-core— model, tools, and streaming pipeline.sharp— image stitching for multi-image prompts.
For install steps, endpoint reference, YOLO/ephemeral details, and script integration, see USAGE.md.
Scan the QR code below to view the repository:
MIT License | Built for high-speed, local-first AI development.