用 Grok Bot 搭一人公司 agent swarm 的 20 个仓库

Gresham @Greshamtrade原推发布于 2026.10.25

在 X 上打开原推

f*cking insane. 95% of people use Grok Bot like a regular chatbot. the other 5% use it to run a $1M one-person company they’ve turned Grok Bot into an agent swarm. each bot owns a specific job, shares files with the others, and hands off finished work. research, outreach, and product updates keep moving on a cloud computer long after the founder closes the laptop 20 GitHub repos to start building your own agent swarm: 01 / BotOps a fleet lifecycle built around a Chief of Staff, domain leads, and temporary workers. Notion holds the tickets, workers claim their tasks, and the lead brings back decisions instead of every bot filling your inbox http://github.com/granda/botops 02 / Bot Org OS a Chief plus Analyst, Product Manager, Architect, and Developer profiles. shared files carry the work between specialists, with source, evidence, and approval checks at each handoff http://github.com/firmasite/bot-org-os 03 / Commander's Intent a Chief of Staff template that gives the fleet one approved mission, spending limits, and decision boundaries. includes skills and routines for coordinating Grok Bot with connected Hermes hosts and Hindsight memory http://github.com/shagghiesuperstar/commanders-intent 04 / Root Agent a coordinator template that scopes the request, selects the team structure, and delegates to specialists or leads. includes independent review and evidence requirements before a task is considered complete http://github.com/MrBekoX/root-agent-skill-framework 05 / GrokBot GTM Agent System a starter team for meeting prep, follow-up drafts, prospect research, and account signals. shared context and voice files keep the work consistent, with a staged path from manual tasks to recurring routines http://github.com/lschwartz-hue/grokbot-gtm-agent-system 06 / TRL Grok Bots a working persona roster covering partnerships, deal handling, briefs, and production. each profile defines its job, what it must avoid, its voice, and when it should wake http://github.com/alytrl/trl-grok-bots 07 / Grok Bot Starter Pack focused profiles for inbound leads, inbox triage, unused subscriptions, and content research. the speed-to-lead kit includes skills, suggested routines, and a first-run checklist, with outbound messages kept as drafts http://github.com/SigiBaes/grok-bot-starter-pack 08 / Grok Bot Profiles separate Spec Writer, Bug Reproducer, PR Producer, and PR Verifier profiles. the verifier works outside the production conversation so the bot building the change doesn’t approve its own work http://github.com/HAEGONG/grok-bot-profiles 09 / Compound Engineering reusable skills for planning, building, reviewing, and recording lessons from engineering work. its documented Grok Bot path uses the Cursor account’s plugin library so Bots can load the same processes http://github.com/EveryInc/compound-engineering-plugin 10 / Elves includes a dedicated Grok Bot skill for long-running work outside code repositories. a plan, survival file, session state, and ledger preserve progress so interrupted work can resume from the last recorded result http://github.com/aigorahub/elves 11 / Hindsight shared memory for the swarm. its Grok Bot integration supports individual banks, a common team bank, tagged handoffs, and notes between recurring runs so another specialist can retrieve earlier findings http://github.com/vectorize-io/hindsight · Grok Bot setup 12 / Kontext a Grok Bot plugin for persistent projects, tasks, documents, and decisions through a hosted MCP service. connected clients can retrieve the same project record instead of rebuilding it in separate conversations http://github.com/TheKontextCo/kontext-grok-plugin 13 / claude-mem has a dedicated Grok Bot integration that watches chat logs and captures useful context across sessions. hosted memory is the default, with a local observer available as an opt-in configuration http://github.com/thedotmack/claude-mem 14 / grokbot-fleet an unofficial MCP management bridge running on the Bot computer. lists Bots, routines, and skills, then routes configuration changes through a Fleet Bot job inbox. profiles can be maintained as YAML files http://github.com/Dahhrk/grokbot-fleet 15 / foreman compares the declared crew with observed Bot activity and surfaces stalled tasks, repeated work, and possible dropped handoffs. reports link back to evidence rather than inventing per-bot token costs http://github.com/Archive228/foreman 16 / Codex Grok MCP an unofficial bridge for bringing named Grok Bots into a Codex workflow. the persistent-Bot path can list, read, wait for, and message specific Bots. that path is experimental, with macOS as the supported starting point http://github.com/Fato07/codex-grok-mcp 17 / grokbot-discord one Discord gateway routes requests to multiple Grok Bots by name, mention, or mapped role. webhook routines wake the selected Bot, and its completed response returns asynchronously. requires a running bridge http://github.com/RudeDude/grokbot-discord 18 / n8n Master Grok Bot connects a Bot to n8n documentation and your own HTTPS-accessible n8n instance. useful for combining Bot decisions with existing automation workflows, starting with a harmless test connection http://github.com/shagghiesuperstar/n8n-master-grok-bot 19 / Overnight a six-Bot production workflow with Chief, Researcher, Director, Render, Editor, and Publisher roles. includes instructions, roster setup, and work records. media generation needs its own connection and budget; publication stops for human approval http://github.com/Archive228/overnight 20 / Omabot a read-only Omarchy desktop widget that shows your Grok Bot roster and which Bots need attention. reads the desktop app’s existing local state so you can spot a waiting teammate without opening every conversation http://github.com/njpatel/omabot pick one coordination framework, then add the specialists and tools your workflow needs. these are building blocks, with different installation requirements, connected services, and running costs give the first crew a concrete delivery: research eight target accounts, prepare tailored outreach drafts, and return a checked package by morning. name who owns the assignment, who produces it, and who checks it the handoff should carry the finished file, its supporting evidence, and the next action. your role is to resolve decisions and approve external actions watch what happens between deliveries. duplicated research, oversized messages, and Bots repeatedly checking on each other can consume the allowance before the useful work arrives save this. build the swarm around a job you want off your desk the article below explains 7 changes for keeping that swarm affordable as it grows, including tighter working context, reusable findings, and limits on failed retries ↓

这条是引用着另一条写的：x.com/greshamtrade/status/2112214741103825314