Scope
The original report links to 22 GitHub projects, verified on 2026-11-04. It labeled the collection as 18 by grouping some projects together. The entries here list them separately while retaining the original A/B/C groups and order within each group; entry numbers are not rankings.
xAI's official page describes named Bots, tool sign-ins, collaboration, persistent context and Routines. Check current official specifications for plans, quotas and isolation details. The applications, directories and clients below provide different layers of functionality; none can be assumed to replace every feature of the others.
Functional layers and evaluation dimensions
| Dimension | Evaluation question | Effect on selection |
|---|---|---|
| Roles and interface | How named Bots, channels and lead/specialist roles divide work | Determines whether to choose a chat workspace, personal Agent or organizational control plane |
| Computers and permissions | Whether work runs on the local machine, a shared computer, a private container or a Git worktree | Workspace separation, file separation and operating-system isolation require separate checks |
| Memory and scheduling | Which state persists, and which resident process runs Routines | Tasks may stop when the desktop app closes, the computer sleeps or the server stops |
| Human approval | Which tools require approval, whether approval can be withdrawn and when a person can take over | Approval cards, policy rules and operation logs cover different scopes |
| Portable configuration | Whether sharing includes prompts, skills, role settings or an entire workflow | Imported recipes still require accounts, computers and tools to be connected again |
A: Named Bots and computer workspaces
This group provides conversational AI coworkers, persistent state and computer operation. Compare execution location and approval methods first, then choose a desktop app, resident server or mobile entry point for the intended use.
Rank 01
milind-soni/OpenMausBotOpenMausBot
Bot workspace driven by local CLIs
Uses signed-in Claude, Codex or Grok CLIs to run named Bots, with channels, computer operation, approval cards and Markdown team packs. Cloud computers, models and connected services may still incur charges; enterprise/ has a separate license from the Apache-2.0 core.
- Local CLI
- Team templates
- Desktop app
Rank 02
elie222/rakazoRakazo
Persistent AI teammates for self-hosted operation
Provides conversations, memory, Routines, voice and peer delegation, with browser, terminal, file and graphical-desktop support. Shared Team Computers and isolated Private computers coexist; choose by use case. The project remains in beta.
- Self-hosted service
- Shared/private computers
- Voice
Rank 03
CopilotKit/OpenBotOpenBot
AG-UI Agents and an adaptable workspace template
Connects AG-UI Agents and provides computer tools, CEL policies, operation logs and human takeover. By default, each user shares one container across Bots, with separate folders and browser profiles per Bot; an Isolated Bot uses its own container.
- AG-UI
- Policy rules
- Template
Rank 04
Anil-matcha/open-duetOpen Duet
AI workspace with a Today view and delegation
Named Agents retain memory, run schedules and receive work delegated by a lead; Today brings approvals, questions and results together. Self-hosting and computer features remain experimental. The README's verified mobile-pairing scope is Android; it does not establish iOS availability.
- Task delegation
- Today
- Mobile pairing
Rank 05
najmuzzaman-mohammad/gawkbotgawkbot
Go workspace with Microapps for viewing work results
Can take over existing local Agent CLIs, provide Routines, sourced knowledge pages and structured data, and generate Microapps for managing work. It uses the Sustainable Use License; the mobile inbox, terminal approvals and Routine packaging remain partly completed or planned.
- Go
- Microapps
- Restricted license
B: Agent runtimes and company orchestration
This group handles personal Agents, development worktrees, delegation across models and multi-Agent governance. Organization charts, budgets and task states do not automatically provide separate browsers or computers; configure those through each project's execution backend.
Rank 01
NousResearch/hermes-agentHermes Agent
Personal Agent with memory, skills and messaging entry points
Provides memory across conversations, history search, skill creation, sub-Agents and a messaging gateway. The README lists six built-in terminal backends: local, Docker, SSH, Modal, Daytona and Vercel Sandbox. Generated skills still require review; do not assume every task produces a reliable skill.
- Personal Agent
- Skills
- Messaging gateway
Rank 02
paperclipai/paperclipPaperclip
Control plane for multi-Agent goals, tasks and governance
Manages external Agents through organizational relationships, goals, tasks, heartbeats, approvals and budgets, with work and spending records. It orchestrates Agents without automatically giving each one a separate computer; spending thresholds are not guaranteed caps on provider bills.
- Organizational orchestration
- Task governance
- Budgets
Rank 03
tinyhumansai/openhumanOpenHuman
Agent harness embeddable in Rust applications
Built around a Rust library, with desktop, terminal and headless modes, plus model, memory, tool, MCP, skill and approval options. The project is labeled early beta and uses GPL-3.0; this review did not retest the original's Agent-capacity and resource-saving figures.
- Rust
- Harness
- GPL-3.0
Rank 04
stablyai/orcaOrca
Git worktrees for managing coding Agents
Displays CLI Agents, split terminals and separate worktrees together in a desktop interface. Design Mode sends page elements and screenshots to an Agent; a mobile companion is also available. Worktrees separate code changes without isolating operating-system permissions.
- Development
- Git worktree
- Design Mode
Rank 05
zeenie-ai/OpenCompanyOpenCompany
AI employees configured through work, memory and skills
Lets lead and specialist roles use tools, browsers, schedules and webhooks according to the work, while retaining knowledge through coaching feedback. The README explicitly says this does not retrain the underlying model; background work still requires the app to run and the computer to remain available.
- Workflow
- Memory and coaching
- Self-hosting
Rank 06
alexgreensh/outsourcererOutsourcerer
Delegation, review and persistent tasks across harnesses
Distributes work across models and harnesses, with advisor reviews, parallel workers, persistent plans, verification and sequential merges. The Grok Bot webhook path is beta and has not completed verification with a real Bot. It uses PolyForm Noncommercial, with a written small-business exception.
- Cross-harness
- Verification workflow
- Noncommercial license
C: Directories, host modifications, clients and domain recipes
Directories help discover recipes; routers depend on official host versions. Clients and domain tools have their own system and account requirements. Before adoption, check whether a tool connects to an existing Grok Bot or needs a separately deployed runtime.
Rank 01
majiayu000/awesome-grok-botAwesome Grok Bot
Directory of public Bot shares and team packs
Organizes x.ai/bot share pages, team packs and use cases by purpose, with a searchable website. Accounts and tools still need to be connected after adding a share; a working link does not establish functional or security validation of the workflow.
- Sharing directory
- Team templates
Rank 02
kydlikebtc/awesome-grokbotawesome-grokbot
Sharing index with provenance and dead-link tracking
Provides bilingual search, source attribution, URL filtering and a dead-link list. It is a resource index without a Grok Bot runtime; check licenses for the directory data, scripts and upstream entries separately.
- Sharing index
- Source attribution
Rank 03
sit-in/awesome-grokbot-playbooksGrok Bot Playbooks
Chinese work recipes and use cases
Collects multi-Bot collaboration, Routines, personal workflows and community tutorials, with markers for the maintainer's verification depth. Read each recipe's inputs, outputs and limits before adapting it to your accounts; the author's testing markers do not mean this page repeated those tests.
- Chinese recipes
- Routine
Rank 04
RongleCat/awesome-grok-botAwesome Grok Bot Resources
Directory of tutorials, tools, comparisons and failures
Collects official documentation, tutorials, practical examples, skills, MCP, comparisons and failure reports, with multilingual editions. It is community resource navigation; check the original articles and their versions rather than assuming every link was retested.
- Resource navigation
- Failure reports
Rank 05
promptadvisers/grokrouterGrokRouter
Experimental model routing for specific official versions
Lets a Bot choose the Codex SDK or OpenRouter while retaining the official conversation, computer and tool-permission architecture; native maintenance inference may still use Grok. The README supports only the exact 0.30.0 and 0.36.0 versions and reviewed host fingerprints. It does not support 0.44.0 or unlisted versions; the license is limited to personal/noncommercial use.
- Experimental routing
- Version limits
Rank 06
abhaysudhir/ungrokungrok
Official CLI sign-in on a remote computer
Attempts to connect an existing subscription to a Grok Bot host through Claude Code or Codex clients while retaining the original app and tools. Fresh installation and follow-up replies to Claude tool results remain unverified; past success does not validate the current installation path or guarantee that Grok usage stops.
- Host modification
- Experimental integration
Rank 07
enderzcx/grok-bot-switchGrok Bot Switch
Provider switching for older cloud-inference hosts
Provides a switching panel and CLI for protocols including OpenAI Chat/Responses and Anthropic Messages. The README explicitly says it cannot switch the main conversation model of newer web Bots, and v0.8.7 rejects those hosts; Claude OAuth is also unsupported.
- Older hosts
- Compatibility limits
Rank 08
codeaashu/grokbot-shimgrokbot-shim
Linux local host and noVNC desktop
Extracts the required runtime from the user's installed official Grok Bot, connecting a local host, configurable model backend and Chrome/XFCE desktop. The project does not include official program files and uses undocumented integration interfaces. The current README covers Linux; updates to the official app may break the integration.
- Linux
- Local runtime
- noVNC
Rank 09
ScriptedAlchemy/grok-bot-cligrok-bot-cli
Terminal management of existing official Bots
Manages Bots, groups, messages and skills using the signed-in desktop app's session and routing. It manages an official workspace without replacing the cloud runtime. Check connectivity with read-only commands first; batch management is not an isolation mechanism.
- CLI
- Existing workspace
Rank 10
Ayuilos/MiffanMiffan
Android AI client with remote computers and approvals
Combines persistent conversations, companion collaboration, permission cards and user-supplied model services in a mobile workspace, with SSH or remote-desktop computer connections. It includes no model and uses AGPL-3.0. It is an independent Android client; visual similarity does not establish compatibility with the official Grok Bot protocol.
- Android
- Remote computers
- AGPL-3.0
Rank 11
galleonlabs/hypergrok-trading-deskHyperGrok Trading Desk
Hyperliquid research, risk and review recipes
Provides seven role definitions, skills and read-only Python market-data tools so Agents can research, review and recheck trades through a shared ticket workflow. Installation does not automatically connect a wallet or authorize trading; this review did not test trading performance or the completeness of approval controls.
- Domain recipes
- Public market data
- Research mode
Setup paths for the main options
These setup paths come from the pinned README revisions and were not executed in this review. Commands containing latest or dynamic installer URLs may still fetch newer versions. Read installer scripts, costs and network-exposure settings first, and use test data.
OpenMausBot
The README's terminal entry point requires Node 24+; follow the prompts to configure Agent CLI sign-in or model access. Source development also requires pnpm. Keep the terminal running: ending the process stops the server while retaining saved settings.
npx openmausbotFor source development, run pnpm install, then run pnpm dev:server, pnpm dev and pnpm dev:desktop in separate terminals. The harness is at 127.0.0.1:8799 and the web app at http://127.0.0.1:5199; keep all three processes running separately.
Rakazo
The public-image installer requires Docker Engine, the Compose plugin, curl and OpenSSL. It does not require cloning the source or installing Node first.
mkdir -p rakazo
cd rakazo
curl -fsSLO https://raw.githubusercontent.com/elie222/rakazo/main/infra/compose/install-images.sh
bash install-images.shOpen http://127.0.0.1:5173, create an account and connect a model. The installer generates or retains .env and enables a local Docker computer by default; choose shared or private computers for the intended use. For public service, follow the project's HTTPS, authentication and deployment guides.
OpenBot
OpenBot requires Docker, Bun 1.4+, a CopilotKit Intelligence project/key and a model key. This local template path uses managed Intelligence; follow the corresponding README steps for self-hosted Intelligence.
git clone https://github.com/CopilotKit/OpenBot.git
cd OpenBot
cp .env.example .env
bun install
npx --yes copilotkit@latest login
npx --yes copilotkit@latest project select
bun scripts/setup-learning.tsAdd required model settings such as OPENAI_API_KEY to .env, then start:
bash scripts/start.shThe app is at http://localhost:3010. The default OPENBOT_SINGLE_USER=true treats requests as the same administrator and is for local setup; configure production sign-in before allowing others to connect. This project is an editable template and beta starting point, without established enterprise-compliance or security certification.
Hermes Agent
Use the README installation entry point for Linux, macOS and WSL2. If the shell has not loaded the command after installation, reload the corresponding shell configuration, then run hermes.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermesChoose the model and credentials during setup. Configure and start hermes gateway only when a messaging entry point is needed; native Windows and Termux have separate installation guides.
Paperclip
The npm entry point requires Node.js 24.11+:
npx paperclipai@latest onboard --yesFollow the prompts to configure the Agent harness and model access. For source development, use pnpm install and pnpm dev, with pnpm 9.15+; also check Rust or prebuilt runner requirements. The development UI/API is at http://localhost:3100. Configure local loopback setup and externally authenticated operation separately.
Architecture comparison of the main workspaces
| Option | Computers and isolation | Models and services | State, scheduling and approvals |
|---|---|---|---|
| Official Grok Bot | The same user's Bots share a persistent cloud computer, files and sign-ins; isolation is per user | Models and plans follow the current official offering | The official page describes persistent context, Routines, collaboration and requests for approval when needed |
| OpenMausBot | Local CLIs, cloud computers or operating-system-supported local control; access depends on configuration | Can reuse CLI sign-ins; models, cloud computers and connected services have their own requirements | Channels, team packs and Permission Broker; exporting configuration does not export every service and signed-in session |
| Rakazo | Shared Team Computers and isolated Private computers are both available | Connects user-supplied model credentials through Pi; select a computer provider separately | Persistent conversations, memory, Routines, history and delegation; validate beta features in the target deployment |
| OpenBot | One container per user by default, with each Bot's folder/profile; Isolated Bot uses a separate container | Requires Intelligence and model configuration; AG-UI Agents connect through a shared gateway | CEL policies, audit trail, Routines and human takeover; gVisor requires the optional COMPUTER_RUNTIME=runsc setting |
| Open Duet | Desktop, self-hosted or separately operated hosted service; computer features are optional and experimental | Users supply models and subscriptions | Today, memory, lead/specialist roles and Routines; a Routine does not approve actions itself |
| gawkbot | Local Agents, per-Bot working directories and allowed tools; cloud use depends on configuration | Requires an existing CLI or model configuration | Microapps, Routines, knowledge pages and approval for external writes; mobile and packaging features have unfinished items |
Adoption limits and verification checklist
| Item | Check before adoption |
|---|---|
| License scope | gawkbot's Sustainable Use license, Outsourcerer's PolyForm Noncommercial license and small-business exception, and GrokRouter's personal/noncommercial terms differ. OpenMausBot's enterprise/ has a separate license, and Miffan uses AGPL-3.0; the collection cannot be treated as one license category permitting unrestricted commercial use |
| Sharing directories | The majiayu000 directory and scripts use CC0/MIT respectively. kydlikebtc data rows may be CC0 or attribution-required CC-BY-4.0, while scripts use MIT. Adopters still need to recheck the accounts, permissions and actions connected to a Bot |
| Version compatibility | GrokRouter, ungrok and Grok Bot Switch depend on different host/client versions. Read each pinned README and check compatibility conditions and recovery methods; a generic one-line installer cannot replace these checks |
| Data and credentials | Local-first does not guarantee that model, connected-service and cloud-computer data remain local. The OpenMausBot CLI README states that API keys may be stored as plaintext in private config.json with Unix permissions 0600; protect the host and backups |
| Costs and capacity | Existing subscriptions remain subject to provider plans, quotas, API charges, computer charges and third-party service fees. Measure high-concurrency, low-cost and caching-savings claims against your own workload |
| Domain actions | Trading, sending email, publishing and deletion require separate explicit authorization. HyperGrok's public market-data entry point supports research; installing a recipe does not authorize wallet access or real trades |
- Not completed:
Check the official README commit, system requirements and license file.
- Not completed:
Create the first workspace with test data containing no production accounts or secrets.
- Not completed:
Explicitly choose shared or private computers and check readable files, domains and tool permissions.
- Not completed:
Use low-risk tasks to test allow, deny, cancel, human takeover and execution logs.
- Not completed:
Close or restart the app and check how memory, Routines and unfinished tasks recover.
- Not completed:
Compare actual model and computer bills before deciding on resident deployment and concurrency.
Next steps by requirement
Local multi-Bot workspace
Compare OpenMausBot, Rakazo and Open Duet first
Compare CLI sign-in, work state, computer modes and human approval using the same low-risk task. Confirm the local workflow before deciding whether to run a resident service.
Custom Agents and policies
Build an evaluation environment from the OpenBot template
Configure Intelligence, models and production sign-in first, then check shared/Isolated computers and CEL rules. Validate actual actions; the default template is not a completed enterprise deployment.
Personal automation across messaging tools
Evaluate Hermes Agent or OpenHuman
Compare execution backends, memory, tools and skill-maintenance needs; measure capacity and performance separately. Check OpenHuman's beta status and GPL terms before adoption.
Multi-Agent governance and development
Choose Paperclip, Orca or Outsourcerer by control layer
Use Paperclip for organization, tasks and budgets; Orca for development worktrees; and Outsourcerer for delegation and review across harnesses. Check how existing Agents connect and what commercial licensing permits.
Existing official Grok Bot workspace
Start with directories, Playbooks and read-only CLI commands
Compare shared configurations and permissions before importing recipes or checking the current state with grok-bot-cli. Evaluate host-modification versions and recovery separately; do not apply older switching methods directly to newer web Bots.
References
- xAI Grok Bot product overview
- milind-soni/OpenMausBot README snapshot
- elie222/rakazo README snapshot
- CopilotKit/OpenBot README snapshot
- Anil-matcha/open-duet README snapshot
- najmuzzaman-mohammad/gawkbot README snapshot
- NousResearch/hermes-agent README snapshot
- paperclipai/paperclip README snapshot
- tinyhumansai/openhuman README snapshot
- stablyai/orca README snapshot
- zeenie-ai/OpenCompany README snapshot
- alexgreensh/outsourcerer README snapshot
- majiayu000/awesome-grok-bot README snapshot
- kydlikebtc/awesome-grokbot README snapshot
- sit-in/awesome-grokbot-playbooks README snapshot
- RongleCat/awesome-grok-bot README snapshot
- promptadvisers/grokrouter README snapshot
- abhaysudhir/ungrok README snapshot
- enderzcx/grok-bot-switch README snapshot
- codeaashu/grokbot-shim README snapshot
- ScriptedAlchemy/grok-bot-cli README snapshot
- Ayuilos/Miffan README snapshot
- galleonlabs/hypergrok-trading-desk README snapshot
- najmuzzaman-mohammad/gawkbot license
- alexgreensh/outsourcerer license
- promptadvisers/grokrouter license
- kydlikebtc/awesome-grokbot license
- Ayuilos/Miffan license