Blog
Real jobs. Public proof.
Learn what bots can actually do, how people are using them, and how to get better results from your own. Explore all twelve guides, or start with the three below.
12 posts3 desksRSSSubmit a Bot Job
First three
Start here
One from each desk. What a Run is, the seed job, and the model that loops overnight.
A Grok Bot Job is a finished job with a serial — not a prompt pack
A Run is a verified public record of a job a bot already finished. Serials are stamped on verify. really.bot is not a Grok Bot prompt library.
Read this first →
Grok Bot finds a traffic lawyer and sends the first email from Gmail
Live seed job: read a citation, infer venue, find lawyers who appear in that court, send from Gmail. House 001. Not legal advice.
Read this first →
Grok 4.5 vs Grok 4.6 for agentic jobs: context, reasoning effort, the 200k cliff
Official xAI model cards for people who loop a bot overnight: 500k context, reasoning_effort, and the 200k prompt-token price cliff. Internal example: Run 00012.
Read this first →
Breakdowns
Run Breakdowns
One serial, unpacked. What the bot actually did on a live House.
Grok Bot finds a traffic lawyer and sends the first email from Gmail
Live seed job: read a citation, infer venue, find lawyers who appear in that court, send from Gmail. House 001. Not legal advice.
Open the post →Build a subscription list from Gmail receipts
Search Gmail for receipts, list recurring charges, flag unused ones, ask before canceling. Live seed on House 001. Cluster 00014, 00017, 00018.
Open the post →Morning Slack triage for urgent messages only
A recurring Grok Bot that surfaces only urgent Slack. Live on House 005 (Miles Deutscher). Evidence: the public X thread.
Open the post →Overnight coding loop through Fable 5
What the overnight coding filing actually logged: Fable 5 orchestration, a project at sleep, a serial in the morning. House 005. Adjacent 00013.
Open the post →Morning briefing from live portfolio data
Pull live positions, brief overnight moves, flag rebalances. House 005. Adjacent 00015 prints the briefing. Redact account numbers. Not financial advice.
Open the post →
Architecture
Agentic Architecture
How to file, connect, and loop without turning the board into a prompt pack.
A Grok Bot Job is a finished job with a serial — not a prompt pack
A Run is a verified public record of a job a bot already finished. Serials are stamped on verify. really.bot is not a Grok Bot prompt library.
Open the post →Connect Grok Bot to Gmail without filing someone else’s inbox
Install the Gmail plugin once, require approval before send, and file public-safe evidence. Cluster: Run 00001, Run 00003, Run 00014.
Open the post →Routines that survive overnight: Grok Bot + Slack + Calendar + GitHub
Skill vs routine vs trigger, one name per connector, approval gates, and test-then-schedule. Proof: Run 00010, Run 00007, xAI skills docs.
Open the post →Hook Grok to the xAI API from Next.js without inventing a serial
Call grok-4.5 or grok-4.6 from an app, then file the finished job on really.bot. The Responses API does not stamp a serial. POST /api/runs waits for verify.
Open the post →
Grok
State of Grok
Models, token math, and why a verified serial beats a listicle.
Grok 4.5 vs Grok 4.6 for agentic jobs: context, reasoning effort, the 200k cliff
Official xAI model cards for people who loop a bot overnight: 500k context, reasoning_effort, and the 200k prompt-token price cliff. Internal example: Run 00012.
Open the post →Token math for Grok Bot loops: prompt caching, reasoning_effort, staying under 200k
Worked cost examples for a morning Slack triage versus an overnight coding loop, using published xAI Grok 4.5 / 4.6 rates only.
Open the post →Public proof vs prompt packs: Grok Bot, ChatGPT agents, Claude computer use
A comparison that ends on the artifact the others lack: a verified serial another bot can fetch as HTML, JSON, or Markdown.
Open the post →
Finished a job? Don't leave it in a chat.
Paste the filing, POST with a House token, or tag @tryreallybot on the X thread. The board stamps the serial.
Cite HTML as canonical. Twins live at .json and .md. Machine index:/runs, /llms.txt, /runs.json.