Aakash GuptaThe OpenClaw Guide no PM is Talking About (Masterclass for AI PMs)
CHAPTERS
- 0:00 – 3:40
OpenClaw goes viral: what it is and why it matters for PM leverage
Aakash and Naman frame OpenClaw’s sudden popularity and the core promise: moving from reactive chat to proactive, autonomous execution. They tease the biggest PM question—how to use it safely and productively—and what will be covered (setup, security, Slack, deployment choices).
- •OpenClaw’s rapid adoption and hype context
- •Proactive agents vs reactive LLM chat
- •PM-centric promise: automation while you sleep
- •Security and deployment questions preview (VPS/Mac Mini/local, Slack/Telegram/Discord)
- 3:40 – 4:59
Two setup paths: one-click hosted install vs full local control
Naman explains the two main ways to get started: a quick “training wheels” one-click install via Emergent versus the more flexible terminal-based local setup. They clarify why the hosted path is easier but restrictive for advanced use cases.
- •Emergent one-click install flow and its limits
- •Why local install enables full freedom (RAM, local models/files)
- •When a quick hosted setup is still useful
- •Mindset: terminal errors are solvable with search/LLMs
- 4:59 – 8:18
Install in the terminal: NPM install and onboarding flow (including gotchas)
They walk through the minimal commands to install OpenClaw and start onboarding, emphasizing that most issues are just environment setup (e.g., Node/NPM). The onboarding forces key identity/personality choices and includes a security warning you must acknowledge.
- •Command: `npm install -g openclaw@latest`
- •Command: `openclaw onboard` and what it generates (e.g., soul.md)
- •Navigation tips: no mouse; use arrow/tab/space/enter
- •Rule of thumb: warnings ok; errors show in red
- 8:18 – 12:12
Connect an LLM provider: API keys, model selection, and performance tradeoffs
Naman demonstrates choosing a model provider (Gemini in this demo) and pasting the API key securely. They discuss model choice tradeoffs: best model vs faster ‘flash’ models depending on end-user latency needs.
- •How to create and store a Gemini API key safely
- •Paste key into onboarding without exposing it
- •Select model (e.g., Gemini Pro vs Flash) based on response-time needs
- •OpenClaw is model-agnostic (swap providers/models later)
- 12:12 – 19:09
Gateway concept + Slack socket mode setup: tokens, permissions, and channel access
They explain the “single gateway” architecture and then do an end-to-end Slack integration walkthrough using Socket Mode. The chapter covers creating a Slack app, generating app/bot tokens, adding scopes, installing to workspace, and the common ‘reinstall’ pitfall.
- •Why OpenClaw uses a single gateway for multiple surfaces (Slack/WhatsApp/etc.)
- •Slack app setup: Socket Mode token (xapp) and Bot token (xoxb)
- •Adding required bot token scopes and what they enable
- •Critical step: reinstall the app after permission changes
- 19:09 – 22:33
Skills vs tools vs hooks: configuring what the agent can do and knows
After Slack setup, they clarify OpenClaw’s extensibility model: tools enable capabilities, skills encode know-how, and hooks run lifecycle TypeScript. They show how you can skip during onboarding and later edit local files directly for faster iteration.
- •Tools vs skills analogy: ‘organs’ vs ‘textbooks’
- •Skills marketplace/community examples (incl. smart home)
- •Hooks explained: lifecycle TypeScript running inside the gateway
- •Practical workflow: skip in onboarding; modify local config files later
- 22:33 – 22:52
Using the TUI and gateway web dashboard as your control center
They “hatch” the bot and introduce the TUI experience plus the local web UI (127.0.0.1) as the command center. This includes how to check gateway status, inspect configuration/secrets, and switch models via config or chat instructions.
- •TUI hatch flow and why OpenClaw asks identity/personality questions
- •Gateway dashboard overview: chat, overview, config, secrets
- •Where to verify connectivity and troubleshoot integrations
- •Model switching via config (or instructing in chat)
- 22:52 – 36:47
Local vs VPS vs separate machine (Mac Mini): reliability and risk tradeoffs
Aakash and Naman address the deployment debate: local execution is safer by default because the agent sleeps when your laptop sleeps, while VPS can run nonstop but raises risk. They also discuss using a dedicated machine (like a Mac Mini) as a middle ground for security and uptime.
- •Two decision factors: usage needs and risk appetite
- •Local runs stop when laptop closes (safer, less autonomous)
- •VPS provides always-on behavior but can be riskiest if misconfigured
- •Dedicated machine can isolate risk while keeping uptime
- 36:47 – 46:33
Use case 1: Turn a Slack channel into a living product knowledge base from docs
They build an AI knowledge base by placing PRDs/FAQs into a local docs folder inside OpenClaw’s workspace, then querying the bot from Slack. The bot reads and updates documents autonomously, demonstrating persistent, editable documentation workflows.
- •OpenClaw workspace: `.openclaw` folder and how to locate it
- •Docs folder pattern: drop PRDs/FAQs/product wiki for retrieval
- •Slack Q&A: bot searches local docs to answer contextual questions
- •Autonomous doc editing: bot appends new FAQs and saves changes
- 46:33 – 53:54
Use case 2: Automated standup summaries via cron jobs (Slack channel scanning)
They configure a recurring standup summary that scans multiple Slack channels, extracts key updates/blockers, and posts a concise brief to a standup channel. They show both manual triggering and cron-based scheduling with configurable prompts and catchphrases.
- •Standup automation: summarize last N hours across channels
- •Cron job creation from plain-English instructions
- •Catchphrase/trigger workflow (e.g., ‘Stand up’)
- •Review and edit scheduled jobs in the gateway ‘cron jobs’ UI
- 53:54 – 1:06:59
Use case 3: Competitive intelligence pipeline (web monitoring + Brave API)
They build a competitive intel automation that monitors competitor sources and posts SWOT-style briefs into a dedicated Slack channel. The chapter includes troubleshooting missing channels, retrieving Slack channel IDs, and adding the Brave Search API key to enable browsing.
- •Create a dedicated ‘competitive intel’ Slack channel and invite the bot
- •Fix common error: channel not found → copy channel ID into config
- •Brave Search API key setup for agent web browsing
- •Result: threat/opportunity insights and persistent tracking over time
- 1:06:59 – 1:12:34
Security & safety: real risks, guardrails, and how to audit your setup
They share cautionary stories (unexpected WhatsApp behavior) and discuss how permissions plus filesystem access can create real security issues. Naman demonstrates asking OpenClaw to audit its own vulnerabilities and generate a remediation plan, emphasizing ongoing ‘security hygiene.’
- •Risks: unrestricted filesystem access + Slack users can prompt exfiltration
- •Default guardrails help, but integrations increase blast radius
- •Run an agent-driven security assessment (and even schedule it weekly)
- •Mitigations: firewall, least-privilege permissions, careful channel access
- 1:12:34 – 1:23:43
Use case 4: Voice of Customer reporting across email + Slack + other sources
They set up a multi-source VoC pipeline that ingests customer feedback from Slack and email (and can extend to reviews, Reddit, Product Hunt, etc.). The automation categorizes themes, tracks trends over time via persistent memory, and posts summaries back to Slack for drill-down.
- •Aggregate feedback sources: support Slack, forwarded emails, reviews/forums
- •Persistent memory enables week-over-week trend analysis
- •Email access via Apple Mail (plus Brave/web sources when needed)
- •Weekly VoC synthesis plus more frequent collection intervals
- 1:23:43 – 1:36:49
Use case 5: Smart bug routing by customer tier (decisioning + logging)
They demonstrate an escalation workflow where bug reports are parsed, mapped to a customer CSV (enterprise/pro/free), and routed to different channels with different urgency. The bot acknowledges in-thread and logs structured entries to a CSV, showing true ‘decision + action’ automation.
- •Parse reporter identity from Slack bug message
- •Match against customer tier list (CSV or a real system like Jira/Asana)
- •Enterprise bugs escalate urgently; free-tier routes lower priority
- •Auto-acknowledge and append a structured log entry to bugs.csv
- 1:36:49 – 1:40:51
OpenClaw vs Claude vs Claude Cowork: autonomy, daemon model, and cost
They compare OpenClaw’s always-on ‘daemon’ and persistent memory to more reactive tools, positioning autonomy as the key differentiator. They also highlight cost flexibility from model-agnostic support (e.g., cheaper open models) versus expensive proprietary APIs for heavy workloads.
- •Claude: powerful but browser-bound and reactive
- •Cowork vs OpenClaw: autonomy and persistence as the main wedge
- •Always-running daemon enables continuous monitoring and memory over months
- •Model choice and cost control (e.g., cheaper models for routine tasks)
