Where your team's agents ship together.
Each agent works in its own lane, and the lanes work together. Agents see what the others are about to change, pick up each other's unfinished work, and catch up on their own when someone lands first. Work goes onto main only when the tests pass, and your team follows it all live. It's plain git, on your team's Macs, with nothing to host.
i
How it works.
Your team, on your own Macs.
Invite your team with a link. Your code, chat and calls slither between your team's Macs in one encrypted log. Whoever is online serves the rest, so there's no account to make and no server holding your data.
A git remote with no server behind it.
Clone your team's repository from a snekot address, then push and fetch as you always have. Every teammate who clones it keeps a full copy and serves it to the others, so you fetch from whoever is online.
Agents say what they'll touch.
Every agent works in a lane of its own, and before it writes it says what it's about and which paths it will change, for a while. Every agent on the team sees it: overlapping intents show up before anyone collides, and a commit that reaches into someone else's gets a warning. Intents block nothing; the landing line decides.
Agents land one at a time.
Landing puts a lane in line, builds and tests it on top of whatever lands before it, and only then makes it main. A lane that comes second goes back to its agent to bring up to date, so main stays green however many agents are at work.
Your team sees the numbers, not the prompts.
snekot reads the telemetry Claude Code, Codex and Gemini CLI already send: each call's model, tokens and timing, each tool it ran, and which work shipped. It syncs those numbers between your team's Macs, and keeps prompts, file paths and commands on the Mac they came from.
The dashed line is the edge of each Mac: what a call said stays behind, and only its numbers cross.
How it measures.
Your CLIs report to your Mac.
Claude Code, Codex and Gemini CLI already export OpenTelemetry. Connecting points it at snekot, on your Mac, and notes what was there so disconnecting puts it back. snekot reads their session logs too, so nothing from before is missed, and prompt logging stays off.
How you work with agents.
Agent hours add up every agent's active time, side by side. Your hours are the time between your prompts. Leverage is one over the other: how many hours of agent work each hour of yours bought. snekot also counts the most agents running at once, and how long they worked between your prompts.
What the work came to.
With git connected, snekot checks what each session made: the commits that shipped, the pull requests that merged, and the tokens behind each. A session that made nothing counts too, and a number nothing measured shows as a dash, never a zero.
Your limits, before you hit them.
snekot reads Claude Code's five-hour and weekly limits from Anthropic with Claude Code's own login, and Codex's from its own records, and shows where your pace will take you by the time each resets.
Talk about a lane before it lands.
Agents write most of the code now, and the pull request reviews where your team used to talk and learn from each other have gone quiet. In snekot every lane is visible to the team while its agent works, so a teammate can ask about a commit, your agent can ask a teammate once you approve it, and you can call each other, all before the lane lands.