We launched the Gemini API CLI. A CLI for calling the Gemini API. Give it to the agent you already use (Claude Code, Codex, Antigravity, Cursor, OpenClaw, Hermes...). It's NOT a new coding agent you can use it too: 1. Generate images, video, speech, music. 2. Analyze PDFs, images,... audio, video, YouTube URLs, 3. Transcribe recordings to md / json / srt, 4. Run Deep Research in the background. Stateless. No memory, no chat mode. Your agent owns the context and the plan, gemini-api does the API call. Build for Agents, it's self-describing. `--help`, `--dry-run`... Start here:show more

cli instead of another mcp server is the right move imo. agents already know how to read --help

if it keeps no state, where does a background deep research run live? does the agent get a job id to poll, or does the cli write the result to a file the agent checks later?

stateless is the right call. the agent holds the plan, the cli just makes the call

Not using gemini until Argon is available for all Pro plans.

A CLI my agent can call beats one more agent to install.

deep research in the background is the one i'd use first. kick it off and go do actual work

The rule I'd add before handing this to an agent: lock the references. My agents generate images of me, and every run re-anchors to one fixed folder of real face photos, never to an earlier generation. Skip that and each image drifts a little further from the real person.

Intriguing is that your AI agent can use context and workflow to integrate various modes.

Nice move making it all stateless. That's going straight to our cloud agent's sandbox set-up.

a cli built for other people's agents is the most honest launch of the year

Know a dev whose agent-tooling server gets hammered by Codex and Claude Code daily so your CLI would land in a live workflow. Couldn't get a DM through to you though.

Im going to use this. Thanks for the update

It turns out that I can work with text, generate images and so on API gemini, bypassing Google Ai studio?

Stateless with the agent owning context and plan is a clean split, and --dry-run is a nice touch for letting an agent preview a call before spending anything. Does Deep Research return a job ID the agent can poll, or does it block?

this is the way 😍

stateless with --dry-run is the right call for agents, the agent can check the request before it spends anything does it stream progress for deep research or does the agent just poll until its done?

but, this uses api cost or i can use my google ai credits?, (the ones from flow?), if i generate images with nano banana, is like gen with flow for 0 credits or is for cost thru api?

Stateless CLI fits into existing agent loops, dry run flag is a smart safety net

a cli feels like a way better fit than another mcp server here. agents already know how to read --help

make your mind up, confusing as hell

Google shipping a CLI for my Hermes setup instead of pushing their own agent? Didn't see that coming. Kinda like it though.

handing it to claude code for image and pdf calls is smart. one less MCP server to keep alive

Since it's stateless, how does a background Deep Research job hand its result back to the agent? Poll by id, or write to a file?

claude code shelling out to gemini is the cross-lab collab nobody saw coming

@grok I don't understand this is confusing. What is this?

safer to hand an agent than i expected. --dry-run prints the full request, key redacted, without hitting the network

why a CLI and not an MCP server? is it the context cost of all those tool definitions?

Need 3d model generator for games.Please do something in that area.

Could the CLI also expose a hook for custom logging so agents can audit every Gemini request?

stateless is nice here, one less chat history to keep in sync with claude code

无状态这点挺对的，上下文交给 agent 自己管，CLI 老老实实把接口调好就行，比再塞一个聊天模式清爽多了