strategy draft

Competitive Landscape

Three ways to place Obelisk. Structural duplicates are rare. The capability envelope gets crowded but narrows back down at depth. And the unbundled point-solution stack shows what a founder would otherwise wire together by hand — credentials, compute, browser, channels, docs, UIs — to approximate what one Scribe already does.

Obelisk
Open-source or managed cousin
Thin layer: builder, SDK, shared runtime
Platform suite
Infra substrate
Vertical specialist

model 1

Structural duplicates

Best when the question is who is actually closest to Obelisk's deployed shape. The lower-left cluster is the thin-agent layer: agent builders and shared runtimes.

Long-running operator Helper or assistant
Zapier Agents

Shared agent builder

Very broad thin-agent layer product: real delegation, but not dedicated long-lived customer runtime.

Zapier says its Agents create specialized teammates that work across 8,000+ apps.

Gumloop

Agent workflow framework

Specialized agents and templates, but still fundamentally a builder on shared infrastructure.

Gumloop markets specialized agents and an AI automation framework for teams.

n8n

AI workflow builder

Strong agent-builder surface with explicit workflow control and optional self-hosting.

n8n says it lets teams build AI agents they can follow, with explicit logic and workflow history.

Goose

Open-source local agent

Another cousin in the self-run open-source layer: local, extensible, autonomous, but not a managed long-lived customer runtime.

goose describes itself as a native open-source AI agent with desktop app, CLI, API, and local execution.

Cloudflare Agents

Stateful agents on Durable Objects

More structurally interesting than most builders: persistent state, scheduling, SQLite, and global deployment without machine ownership.

Cloudflare says each agent runs on a Durable Object with SQL storage, WebSockets, scheduling, and no infrastructure to manage.

Pi

Open-source coding agent

Interesting open-source cousin: thinner than Obelisk, closer to the terminal-agent layer, with hints of future managed enterprise/cloud packaging.

The pi package describes an interactive coding agent CLI and standalone OpenAI-compatible agent, while the maintainer notes some enterprise features and cloud infrastructure will be proprietary.

Anthropic Managed Agents

Cloud-hosted managed agent runtime

More relevant than the SDK here: Anthropic runs the sandboxing, state, permissions, and long-running sessions for you, but it is still Anthropic-hosted rather than customer-owned runtime.

Anthropic says Managed Agents is a suite of composable APIs for building and deploying cloud-hosted agents with secure sandboxing, long-running sessions, and orchestration handled for you.

Perplexity Computer

Shared digital worker

Credible operator framing with scheduling, persistent memory, connectors, and an isolated sandbox, but the runtime is still Perplexity-hosted rather than a dedicated customer machine.

Perplexity says Computer is an independent digital worker with app connectors, persistent memory, scheduling, and a secure isolated sandbox.

Polsia

Company-operating agent platform

High-autonomy company OS framing with daily coordinated cycles, but not a general-purpose agent running inside a dedicated customer VM.

Polsia describes itself as an autonomous AI system that runs entire companies through coordinated AI agents that plan, code, and market daily while founders sleep.

Obelisk

Dedicated VM Scribes

Closest to the far edge: managed, isolated, long-lived agent runtime with a control plane.

Obelisk docs describe a control plane for Scribes running on dedicated VMs.

Source: internal Obelisk docs
OpenClaw

Self-hosted gateway

Persistent and agentic, but self-hosted and framed as a personal assistant gateway.

OpenClaw describes itself as a self-hosted gateway connecting channels to an always-available AI assistant.

AgentClaw

Managed OpenClaw hosting

Probably the nearest small-company overlap: OpenClaw on managed infrastructure, 24/7.

AgentClaw says it runs OpenClaw agents on managed infrastructure with uptime monitoring.

MyClaw

Managed OpenClaw cloud

A direct managed-hosting product for OpenClaw: private instance, always-on, updates and maintenance handled for you.

MyClaw says it is managed cloud hosting for OpenClaw with a private always-on instance and no setup.

OpenClaw Host

Dedicated container hosting

Similar on persistence and isolation, but still a hosting wrapper around OpenClaw.

OpenClaw Host markets dedicated cloud containers, isolated pods, and 24/7 uptime for OpenClaw.

SnappyClaw

Fully managed OpenClaw

Another managed cousin: OpenClaw paired with hosted infrastructure and model choice.

SnappyClaw says it gives you a fully managed OpenClaw instance paired with the LLM of your choice.

Clawhost

One-click OpenClaw hosting

Same pattern again: productized OpenClaw hosting that removes the self-hosting burden.

Clawhost markets one-click hosting for OpenClaw with no servers or setup.

Factory

Agent-native dev runtime

Not a duplicate, but notable because it packages long-running delegated agents around software work.

Factory positions Droids as software development agents embedded across IDE, web, CLI, and CI.

Browserbase

Browser agent infrastructure

Important adjacent layer. Strong substrate, weak finished-worker positioning.

Browserbase says it is the platform to build and deploy agents that browse and interact with the web.

CrewAI

Multi-agent framework + AMP

Serious orchestration and managed deployment, but still a framework/platform sale where the customer assembles agents, flows, and tools rather than hiring a preconfigured operator.

CrewAI AMP says it extends the open-source framework with managed infrastructure to deploy, monitor, and scale crews and agents in production.

Manus

Cloud browser agent

High agent feel, but centered on a cloud browser environment rather than a provisioned machine.

Manus documents a cloud browser that can log into accounts and act across the web.

ChatGPT agent

Shared browser agent

Strong perceived autonomy, but no dedicated customer-owned long-lived runtime.

OpenAI describes Operator, now folded into ChatGPT agent mode, as an agent using its own browser to perform tasks.

Shared SaaS or browser runtime Dedicated customer runtime

model 2

Capability envelope

Best when the real question is how much work the agent can actually do. Obelisk's dedicated VM, Linux surface, and plugin model create a much broader and deeper work envelope than workflow tools, but specialists still market more cleanly inside a single function.

Deep job ownership Single tasks or workflows
n8n

Inspectable AI workflows

Model: Execution-based cloud plan; free self-host option

Can orchestrate plenty of logic, but still slices work into explicit automations more than owned jobs.

n8n emphasizes traceable reasoning, human-in-the-loop controls, and workflow history.

Relay.app

Predictable AI workflows

Model: Tiered SaaS + bundled AI credits

Useful workflow automation, but narrow on both work surface and ownership depth.

Relay says AI steps fit into reliable, inspectable workflows with full task history.

Gumloop

AI automation framework

Model: Credit-based SaaS with shared org credit pool

Strong automation framework, but the agent is still mostly bounded by workflow design.

Gumloop markets centralized controls, MCP logging, and cloud-or-your-VPC deployment.

Copilot Studio

Enterprise agent builder

Model: Credits pack, pre-purchase, or pay-as-you-go

Broader enterprise surface than pure workflow tools, but still mainly a builder/platform sale.

Microsoft says Copilot Studio lets you create agents with natural language or a graphical interface and supports autonomous agents.

Zapier Agents

AI teammates across apps

Model: Activities/month on free or Pro tier

Broad app surface, but still shallow relative to systems that own longer-running work.

Zapier says its Agents create specialized teammates that work across 8,000+ apps.

Agentforce

CRM-native agent platform

Model: Flex credits, conversations, or user licenses

Can own meaningful work, but mostly inside the Salesforce operating surface.

Salesforce positions Agentforce as the AI agent platform connected to data and tools via MCP.

UiPath

Agentic automation platform

Model: Basic SaaS tier; enterprise agents on standard contract

Broad process automation platform, but still tends to decompose work into orchestrated flows and roles.

UiPath describes agentic orchestration across AI agents, robots, and people.

ServiceNow

Enterprise AI agents

Model: Enterprise package + action or assist consumption

Broad internal enterprise reach, but still strongest where the work stays inside the ServiceNow system of record.

ServiceNow markets AI agents, AI Agent Fabric, AI Agent Orchestrator, and prebuilt agents across functions.

Moveworks

Enterprise AI agent builder

Model: Enterprise contract

Covers a broad employee surface, but with less deep runtime control than Obelisk.

Moveworks positions AI Agent Studio as a way to build enterprise-ready AI agents for all employees.

Aisera

Enterprise AI agents

Model: Enterprise contract

Can go deep, but mostly within IT and support functions rather than a general operator surface.

Aisera highlights AI agents for IT support, software provisioning, procurement, and troubleshooting.

Lindy

AI employee for work life

Model: Tiered monthly plans + voice usage add-ons

General admin worker framing, but not the same systems depth as a dedicated VM operator.

Lindy documents AI-powered workflows and positions itself as the AI that runs your work life.

Ema

Universal AI employees

Model: Enterprise suite / AI employee contract

Strongest broad worker framing among enterprise suites, but still less infrastructure-heavy than Obelisk.

Ema explicitly markets universal AI employees and an AI Employee Builder.

HighLevel AI Employee

AI business operating system

Model: Unlimited plan at $97/month per sub-account

This is the clearest small-business-targeted adjacent offer: broad across calls, chats, funnels, content, and workflows, but still much shallower than a dedicated runtime that can own arbitrary work.

HighLevel calls AI Employee 'The AI Business Operating System for doers' and says it is built for agencies and small business owners.

Polsia

AI company OS

Model: Web SaaS with low-price plans listed publicly

One of the clearest company-operating-system framings: broad business surface and high autonomy, but still more shared SaaS orchestration than deep machine ownership.

Polsia describes itself as an autonomous AI system that runs entire companies through coordinated AI agents and executes development, growth, and operations cycles.

11x

Revenue digital workers

Deep ownership inside one function, but not a broad work surface.

11x says it builds autonomous digital workers focused on revenue functions.

Artisan

AI employees for outbound

Another specialist worker sale: deep in outbound, narrow outside it.

Artisan says it is building AI employees, starting with outbound sales.

Harvey

Legal specialist

Very strong depth in legal work, but that depth does not translate into a broad operating surface.

Harvey markets citation-backed legal research and workflow execution.

Hebbia

Research specialist

Deep research ownership in finance and diligence, but still specialist-first rather than general operator.

Hebbia markets a multi-agent research system for finance, banking, and law.

Glean

Knowledge + agent layer

Model: Enterprise Flex credits per agent run

Broad enterprise surface area, but less depth of owned execution than worker-style products.

Glean launched horizontal enterprise agents and provides an agent builder and library.

Factory

Software engineering agents

Model: Enterprise contract

Deep ownership in software production, but still a specialist work surface centered on engineering.

Factory sells software development agents that handle refactors, incidents, and migrations.

Obelisk

Hire a Scribe

Model: Usage-based labor model: hourly active work + monthly minimum

The core pitch: general-purpose operators with unusually deep execution because the runtime owns a full machine, OS tools, browser, and plugins.

Obelisk positions Scribes as autonomous AI coworkers running on dedicated VMs with a control plane.

Source: internal Obelisk docs
Narrow specialist or function Broad work surface

model 3

The unbundled stack

Best when the question is what a founder would have to wire together by hand to replicate what one Scribe does. The lower-left is the component shelf — credentials, compute, browser, messaging, docs, hosted UIs — each strong at a single capability. Hermes Agent sits mid-frame as the closest open-source framework for persistent agents. Obelisk is the far corner: the whole stack, preconfigured, under one control plane.

Arrives preconfigured User assembles
OneCLI

Credential vault proxy

Model: Open source, free

One capability: encrypted secrets and authenticated API calls for agents. Obelisk ships this inside the typed plugin surface — short-lived tokens, every action logged, no raw keys exposed.

OneCLI markets itself as a local proxy and encrypted credential vault so agents can make authenticated API calls without directly handling raw secrets.

Open Browser

Browser automation framework

Model: Open source, free

TypeScript framework for agent browsing — click, type, extract. A Scribe has this built in alongside shell, plugins, and a full Linux surface.

Open Browser is an open-source TypeScript framework for autonomous web browsing with multi-provider support and sandboxed execution.

Kapso

WhatsApp layer for agents

Model: Paid SDK + API

Single channel. Obelisk Scribes already own email and SMS identity — WhatsApp is another plugin, not a separate product.

Kapso is a developer-focused layer on top of the WhatsApp Cloud API with storage, conversation management, and TypeScript SDK.

Agent Computer

Persistent VM over SSH

Model: Paid per workspace

Literally the substrate Obelisk sits on top of: provisioned Ubuntu boxes for agents. No control plane, no own inbox, no plugins, no audit trail. The VM and nothing else.

Agent Computer offers persistent cloud computers AI agents access over SSH, with Ubuntu sandboxes and persistent storage.

Proof

Agent-aware doc editor

Model: Freemium

Collaborative markdown with provenance tracking for agent edits. A useful surface, but one channel of work. Obelisk Scribes already operate inside Docs, Sheets, and Drive via typed plugins.

Proof is a collaborative document editor with provenance tracking, APIs for edits and comments, and an open-source SDK.

gui.new

Agent-hosted HTML

Model: Freemium

Turns agent output into shareable URLs. A Scribe on a VM can host, ship, or email its own artifacts — this is a feature, not a product layer.

gui.new converts agent-generated HTML into hosted, shareable URLs via a single API call, with multiplayer collaboration.

Hermes Agent

OSS persistent-agent framework

Model: Open source, free

The closest framework cousin on this map. Persistent memory, subagents, Slack/Telegram messaging — but a framework you assemble and host, not a managed operator with a control plane and an SLA.

Hermes Agent (Nous Research) is an open-source platform for persistent agents with long-term memory, reusable skills, subagent delegation, and messaging integrations.

Microsoft AutoGen

OSS agent framework

Model: Open source, free

Historically influential in multi-agent tooling, but still a framework you assemble and host yourself, and Microsoft now steers new projects toward Agent Framework.

The AutoGen GitHub repo describes it as a programming framework for agentic AI and directs new projects to Microsoft Agent Framework for the latest multi-agent capabilities with long-term support.

AgentMail

Email inbox API for agents

Model: Paid API; enterprise tier

One channel: an API that gives agents a real two-way inbox with threading, search, and reply. Obelisk Scribes already own their email identity — this is a component that would sit on the shelf next to Kapso for WhatsApp.

AgentMail (agentmail.to) markets an API-first email platform that gives AI agents their own inboxes with full send, receive, thread, and search capabilities.

OpenClaw

Self-hosted AI gateway

Model: Open source, free

General-purpose channel connector you assemble: Slack, Discord, iMessage, Telegram, web browsing, computer control — all routed through a self-hosted instance. Obelisk ships the equivalent across email, SMS, and plugins without standing up any infrastructure.

OpenClaw is an open-source self-hosted personal AI assistant gateway connecting messaging channels to local or remote LLMs, with web browsing and computer control.

Claude Cowork

Desktop agentic assistant

Anthropic's own agentic product: runs multi-step tasks across local files and apps without step-by-step prompting. Broad work surface, but it is shared infrastructure — you are renting a session on Anthropic's cloud, not owning a machine.

Claude Cowork is a desktop-based AI system that completes autonomous multi-step workflows across files and applications, built around outcomes rather than prompts.

Claude Code (DIY)

Code-agent CLI on a rented VM

Model: API cost + VM cost

The honest technical substitute. Wire Claude Code onto a rented box and bolt on every other item on this map yourself. Capable if assembled — but assembly, maintenance, and drift are on you.

Claude Code ships as a CLI coding agent that can run persistently on a user-provisioned VM.

Obelisk

One Scribe, whole stack

Model: Usage-based labor: hourly active work + monthly minimum

Every item on this map, rolled up. Credentials, compute, browser, shell, channels, plugins, hosted UIs, docs, audit — one control plane, own inbox, own schedule. You hire, you don't assemble.

Obelisk positions Scribes as autonomous operators on dedicated VMs with typed plugins, own email and SMS, scheduled tasks, and full audit trail.

Source: internal Obelisk docs
Narrow component General operator
Research log
11x 11x says it builds autonomous digital workers focused on revenue functions. Source
Agent Computer Agent Computer offers persistent cloud computers AI agents access over SSH, with Ubuntu sandboxes and persistent storage. Source
AgentClaw AgentClaw says it runs OpenClaw agents on managed infrastructure with uptime monitoring. Source
AgentMail AgentMail (agentmail.to) markets an API-first email platform that gives AI agents their own inboxes with full send, receive, thread, and search capabilities. Source
Agentforce Salesforce positions Agentforce as the AI agent platform connected to data and tools via MCP. Source
Aisera Aisera highlights AI agents for IT support, software provisioning, procurement, and troubleshooting. Source
Anthropic Managed Agents Anthropic says Managed Agents is a suite of composable APIs for building and deploying cloud-hosted agents with secure sandboxing, long-running sessions, and orchestration handled for you. Source
Artisan Artisan says it is building AI employees, starting with outbound sales. Source
Browserbase Browserbase says it is the platform to build and deploy agents that browse and interact with the web. Source
ChatGPT agent OpenAI describes Operator, now folded into ChatGPT agent mode, as an agent using its own browser to perform tasks. Source
Claude Code (DIY) Claude Code ships as a CLI coding agent that can run persistently on a user-provisioned VM. Source
Claude Cowork Claude Cowork is a desktop-based AI system that completes autonomous multi-step workflows across files and applications, built around outcomes rather than prompts. Source
Clawhost Clawhost markets one-click hosting for OpenClaw with no servers or setup. Source
Cloudflare Agents Cloudflare says each agent runs on a Durable Object with SQL storage, WebSockets, scheduling, and no infrastructure to manage. Source
Copilot Studio Microsoft says Copilot Studio lets you create agents with natural language or a graphical interface and supports autonomous agents. Source
CrewAI CrewAI AMP says it extends the open-source framework with managed infrastructure to deploy, monitor, and scale crews and agents in production. Source
Ema Ema explicitly markets universal AI employees and an AI Employee Builder. Source
Factory Factory positions Droids as software development agents embedded across IDE, web, CLI, and CI. Source
Glean Glean launched horizontal enterprise agents and provides an agent builder and library. Source
Goose goose describes itself as a native open-source AI agent with desktop app, CLI, API, and local execution. Source
Gumloop Gumloop markets specialized agents and an AI automation framework for teams. Source
Harvey Harvey markets citation-backed legal research and workflow execution. Source
Hebbia Hebbia markets a multi-agent research system for finance, banking, and law. Source
Hermes Agent Hermes Agent (Nous Research) is an open-source platform for persistent agents with long-term memory, reusable skills, subagent delegation, and messaging integrations. Source
HighLevel AI Employee HighLevel calls AI Employee 'The AI Business Operating System for doers' and says it is built for agencies and small business owners. Source
Kapso Kapso is a developer-focused layer on top of the WhatsApp Cloud API with storage, conversation management, and TypeScript SDK. Source
Lindy Lindy documents AI-powered workflows and positions itself as the AI that runs your work life. Source
Manus Manus documents a cloud browser that can log into accounts and act across the web. Source
Microsoft AutoGen The AutoGen GitHub repo describes it as a programming framework for agentic AI and directs new projects to Microsoft Agent Framework for the latest multi-agent capabilities with long-term support. Source
Moveworks Moveworks positions AI Agent Studio as a way to build enterprise-ready AI agents for all employees. Source
MyClaw MyClaw says it is managed cloud hosting for OpenClaw with a private always-on instance and no setup. Source
Obelisk Obelisk docs describe a control plane for Scribes running on dedicated VMs. Source: internal Obelisk docs
OneCLI OneCLI markets itself as a local proxy and encrypted credential vault so agents can make authenticated API calls without directly handling raw secrets. Source
Open Browser Open Browser is an open-source TypeScript framework for autonomous web browsing with multi-provider support and sandboxed execution. Source
OpenClaw OpenClaw describes itself as a self-hosted gateway connecting channels to an always-available AI assistant. Source
OpenClaw Host OpenClaw Host markets dedicated cloud containers, isolated pods, and 24/7 uptime for OpenClaw. Source
Perplexity Computer Perplexity says Computer is an independent digital worker with app connectors, persistent memory, scheduling, and a secure isolated sandbox. Source
Pi The pi package describes an interactive coding agent CLI and standalone OpenAI-compatible agent, while the maintainer notes some enterprise features and cloud infrastructure will be proprietary. Source
Polsia Polsia describes itself as an autonomous AI system that runs entire companies through coordinated AI agents that plan, code, and market daily while founders sleep. Source
Proof Proof is a collaborative document editor with provenance tracking, APIs for edits and comments, and an open-source SDK. Source
Relay.app Relay says AI steps fit into reliable, inspectable workflows with full task history. Source
ServiceNow ServiceNow markets AI agents, AI Agent Fabric, AI Agent Orchestrator, and prebuilt agents across functions. Source
SnappyClaw SnappyClaw says it gives you a fully managed OpenClaw instance paired with the LLM of your choice. Source
UiPath UiPath describes agentic orchestration across AI agents, robots, and people. Source
Zapier Agents Zapier says its Agents create specialized teammates that work across 8,000+ apps. Source
gui.new gui.new converts agent-generated HTML into hosted, shareable URLs via a single API call, with multiplayer collaboration. Source
n8n n8n says it lets teams build AI agents they can follow, with explicit logic and workflow history. Source