Pokee AI
An execution-oriented Agent platform built around isolated cloud VMs, OAuth credential boundaries, and credit-based usage, offering PokeeClaw, Skills, Playbooks, Official Agents, and enterprise deployment options.
Agents run in isolated cloud VMs where they can install packages, retain files, call external tools, and place high-risk write actions behind human approval.
Who it is for: Individuals and small teams, e-commerce and growth teams, product teams needing Agent APIs, and data-sensitive organizations requiring controlled deployment
Positioning
Pokee AI combines a managed Agent workspace, installable capabilities, and packaged Agent services to handle cross-tool execution, file delivery, and enterprise data boundaries.
- The workspace includes Dashboard, PokeeClaw, Skills Marketplace, Playbook, and Official Agents such as Deep Research and the E-commerce Strategy Agent.
- Skills Marketplace contains 54 Skills and shows 12 by default; Playbook is a separate prompt and workflow template library, managed independently from Skill packages.
- Pokee-Isaac 28B is still marked Coming Soon on the homepage; Customer VPC follows an enterprise sales and custom-delivery path rather than direct self-service activation.
Core features
- PokeeClaw isolated cloud workspace: Each user receives an isolated cloud VM with 1 CPU and 4GB RAM, plus Python, Node, Git, and system-package installation. Free storage is 1GB and paid storage is 5GB; files and packages persist across sessions, while files are cleared after 30 days of inactivity.
- OAuth integrations and human approval: The authorization layer connects services such as Gmail, LinkedIn, Jira, and Shopify. Pokee says credentials are injected by a runtime proxy rather than exposed to model context, while external or high-risk write actions require human approval.
- Skills Marketplace: The marketplace contains 54 Skills and shows 12 by default. Official examples cover social scheduling, bug triage, competitor monitoring, and research, while community examples cover job search, contract analysis, and trip planning. Installation requires acceptance of the terms of service.
- Playbook template library: More than 20 one-click prompt and workflow templates cover presentations, cold email, lead research, competitor pricing, SEO, content calendars, and market research. Playbooks and Skill packages are separate mechanisms.
- POKEE.md and MEMORY.md: PokeeClaw stores persistent instructions in a user-editable POKEE.md file. The enterprise runtime uses workspace/.memory/MEMORY.md and .pokee/*.md for memory and persona configuration.
- Scheduled tasks and background execution: Scheduled tasks keep running after the page closes. Any number of active schedules share a 10-credits-per-hour sandbox charge, which stops only after all scheduled tasks are disabled.
- Official Agents: Deep Research uses asynchronous submission and polling and is priced at 60 input and 400 output credits per million tokens. The E-commerce Strategy Agent costs 8,000 credits per report and delivers a strategy PDF and presentation.
- Pokee-Isaac and enterprise deployment: Pokee promotes a 28B Pokee-Isaac model with a 10M context window and customer-boundary execution, plus Pokee Managed dedicated tenancy or Customer VPC deployment. Isaac remains closed and marked Coming Soon, while Customer VPC is a custom delivery path.
Target users
- Individuals and small teams: Use Gmail, LinkedIn, Jira, Shopify, and other connectors for administrative, content, and growth work while retaining files and the task environment.
- E-commerce and growth teams: Produce recurring promotion strategies, competitor research, content schedules, and deliverable PDFs or presentations.
- Product teams embedding Agents: Use Deep Research or enterprise APIs to add long-running research, file output, and usage tracking to their own products.
- Data-sensitive enterprises: Require dedicated tenancy, Customer VPC deployment, credential isolation, human approval, and custom Skill delivery.
Pricing
- Free · $0 / month: 500 credits · 1GB — All major entry points, limited usage
- Lite · $19.99 / month: 2,000 credits · 5GB — Light individual use
- Pro · $49.99 / month: 6,000 credits · 5GB — Frequent individual use
- Ultra · $199.99 / month: 40,000 credits · 5GB — High-volume individual use
- Enterprise · From $2,000–$10,000 / month: 300,000–1,920,000 pooled credits — 100–600 users; higher tiers add API access, audit, SLA, and custom integrations
Verdict
Pokee has built a usable, paid execution-oriented Agent workspace whose strengths are an isolated runtime, credential boundaries, and several reusable task entry points. Isaac remains Coming Soon and Customer VPC follows a custom enterprise delivery path, while Skill governance, failure recovery, credit transparency, and independent user evidence remain the main constraints.
- Strength: PokeeClaw brings runtime, files, external tools, background tasks, and human approval into one workspace, with Skills, Playbooks, and Official Agents providing multiple layers of reuse.
- Limitation: The product narrative is ahead of self-service delivery, runtime reliability and marketplace governance have visible gaps, and independent user and operating evidence remains limited.
How it delivers value
- Task submission: Users submit work through natural language, OAuth connections, a Playbook, or an installed Skill. → Cross-tool work can begin without first constructing a complete workflow.
- Task understanding and plan generation: The system combines its model, POKEE.md, workspace files, and a Skill or Playbook to organize execution steps. → A natural-language goal becomes a traceable execution plan and sequence of tool calls.
- Execution and result delivery: The Agent runs code in an isolated VM, calls external tools, and generates files; external and high-risk writes require human approval. → Users receive email, documents, spreadsheets, presentations, or completed cross-system actions.
- Context retention and reuse: Files, packages, POKEE.md, MEMORY.md, and scheduled tasks remain available in later sessions. → Repeated work can reuse the existing environment and task method.
Competition
Pokee sits between managed Agent workspaces, cross-tool automation, and enterprise Agent runtimes. The key comparison dimensions are execution environment, connector depth, human approval, failure recovery, data boundaries, and cost per task.
- vs Lindy / Zapier Agents: Wins: Each user receives an isolated VM that can install packages, retain files, and run background tasks. Loses: Workflow visualization, template maturity, connector governance, and failure debugging remain behind mature automation platforms.
- vs Local OpenClaw / self-hosted sandboxes: Wins: Managed onboarding, OAuth, credential boundaries, and human approval reduce the operational burden of running the environment yourself. Loses: The 1 CPU / 4GB VM, credit metering, and device exclusivity constrain capacity and control; local deployments let users define the data boundary directly.
- vs OpenAI Deep Research / Claude Cowork: Wins: Pokee puts research and file output inside a persistent workspace and can write back to authorized business systems. Loses: Its foundation-model brand, independent user evidence, and product maturity are weaker, while Isaac is not yet available to ordinary users.
Risks
- Narrative versus self-service delivery: The homepage emphasizes 10M-context Isaac and customer-boundary deployment, while the core self-service products today are credit-metered PokeeClaw and Official Agents.
- Credit transparency: Scheduled tasks, Deep Research, and e-commerce reports have explicit prices, while ordinary PokeeClaw action-level usage is not disclosed.
- Skill governance and security: A marketplace listing contained script-injection content while Agents also have shell and OAuth access, making listing review and permission controls material risks.
- Runtime and approval reliability: Device exclusivity and human approval for high-risk writes interrupt the automated execution chain, making clear recovery, retry, and status feedback important.
- Scope of vendor-reported metrics: Isaac benchmarks and time-saving examples are vendor-reported, while SOC 2 Type I covers a defined attestation scope rather than every production or compliance requirement.
- Limited scale and user evidence: The public community and long-term user sample are thin, while active customers, retention, revenue, and task success are not disclosed.
Company
- Stage: PokeeClaw and Official Agents are available and paid; Pokee-Isaac remains marked Coming Soon; enterprise deployment follows a demo and custom-delivery process.
- Funding: In July 2025, Point72 Ventures led an approximately $12 million seed round with participation from Qualcomm Ventures, Samsung NEXT, and others.
- Founded: 2024
User voices
- Positive: A vendor-published customer case says a task involving multiple login layers and database writing fell from roughly two hours to 20 minutes; the example has limited representativeness.
- Positive: One Reddit user found Pokee faster with similar output quality in a small research comparison while noting that the test was not rigorous.
- Positive: A visible Product Hunt review recognized rapid multi-integration workflow creation and API deployment.
- Negative: The same Reddit comparison said both products missed recent information, so a single result should not be generalized into stable quality.
- Negative: The visible Product Hunt review requested workflow versioning.
- Requested: Public independent evidence is largely limited to one Reddit comparison and one visible Product Hunt review, so it should not be generalized into satisfaction, retention, or task-success claims.
- Requested: Isaac benchmarks and SOC 2 Type I are disclosed in official materials; enterprise buyers should evaluate the deployment scope against their own technical and compliance requirements.
Sources
Updated 2026-08-18