Phala

Our next dstack Community Call is pulling up this Thursday! 🦔 🕐 1:30PM EST 📍 https://luma.com/0di7bo5p Agenda includes: • dstack roadmap update from Phala • Lightning talks from @near_ai & @AskVenice • Live demos: k8s & HA database on dstack • Open Q&A with builders

Most OpenClaw agents ask for your trust. Clawdi doesn’t. It runs in Phala's TEEs, so your data stays protected at the hardware level. Not even Phala can see it. OpenClaw power. 3-min setup. Enterprise-grade privacy. $29/month.

March has been a strong month for Phala, with over 1B tokens processed daily on OpenRouter alone. https://lnkd.in/gAry6M5D

🤝 Phala x Venice AI We’re proud to power Venice AI’s new TEE inference mode: → Your prompts run inside TEE enclaves → A cryptographic attestation proves the model is untampered → No one, not even Phala can see your data Private AI, but verifiable. Details: https://lnkd.in/gW4TEcgQ

TELUS and Fortanix just launched a Confidential AI solution at NVIDIA GTC 2026, with cryptographic proof of protection for sensitive AI workloads on sovereign infrastructure. The market is speaking: verifiable, hardware-backed AI security is becoming the standard. https://lnkd.in/ghaG8K_z

We benchmarked OpenClaw across 8 real tasks + a 5-turn coding conversation. Big takeaway: OpenClaw usually isn't expensive because it writes too much. It's expensive because it keeps re-reading previous context. The worst offender wasn't coding. It was web fetching. - Fetch + summarize a web page: $0.180 - Write a complex TypeScript cache system with tests: $0.033 Why? Messy pages dump huge amounts of raw HTML into context. Another surprising result: - Images were cheap (~$0.011) - Output was only 1-6% of total tokens - There's an ~8k token baseline on every request So the real cost driver usually isn't generation. It's context read. And the real cost explosion comes from multi-turn chat. In our benchmark, a realistic 5-turn coding conversation cost 13.3x the first turn. And running cron jobs in isolated sessions cut monthly cost by ~37% vs using the main chat. Cron work deserves its own rule: if a task runs on a schedule, keep it isolated unless it truly needs your main chat history. Otherwise every health check, summary, or periodic task keeps paying to reload context it doesn't need. If you're trying to cut OpenClaw cost: - start fresh sessions for new subtasks - use isolated mode for recurring jobs - be careful with content-heavy web fetches - optimize what the model reads, not just what it writes Full benchmark: https://lnkd.in/g5dGiZKi

Phala is partnering with Intel on Trust Authority. By integrating Intel Trust Authority (ITA), an independent, third-party appraiser that sits outside Phala's infrastructure entirely, Phala is advancing its attestation capabilities for its Confidential AI infrastructure. Read on: https://lnkd.in/gdyrF3ut

OpenClaw dropped and everyone lost their minds. An AI assistant that remembers. That acts. That automates your entire workflow. Then people tried to set it up. Most hit a wall. Even for experienced devs. Docker. YAML. OAuth flows. API keys. A dependency stack that breaks in creative ways. And once it's running? You're on maintenance duty indefinitely. Local compute alone can run $100+/day. And it doesn't end at setup. OpenClaw was designed for a trusted local machine. The moment you browse the web or install a community skill, that assumption breaks. There are already writeups showing full agent takeover from browser tabs. So here's the irony: The people who could benefit the most from OpenClaw — busy professionals who has fifteen tabs open and three ongoing projects or operations leads whose workflow are tedious and repetitive — are usually the ones least likely to have the time and technical bandwidth to wrestle with it. The Clawdi team wants to close this gap and built Clawdi on top of Phala's infra. The result: a fully functional OpenClaw agent in the confidential compute cloud. Ready in 3 mins. No terminal. No config files. No Docker. No YAML. No Maintenance needed. $29/month flat. You get the full power of OpenClaw: Full web search. Browser automation. File editing. Terminal access. The complete skills system. The Clawdi team manages the model integrations, API keys, and infra. And you get to set it up in under 3 mins. Every Clawdi agent runs in its own TEE. The agent gets a sandboxed file system and a sandboxed browser. It can process your documents, browse the web, and execute tasks, without touching your local machine. Plus the cryptographic attestation you can verify yourself. 🦞 Use case 1: Product Manager A PM shares a screenshot of user feedback. Clawdi reads it, creates a structured ticket, assigns it to the right engineer. No tool-switching. No manual triage. Human completely out of the loop. 🦞 Use case 2: Growth & Content Ops Clawdi scans a target X timeline, reads the thread context, and drafts account-specific replies — tailored in tone, ready to post. You review. You approve. Done. The thinking is handled. Humans just make the call. 🦞 Use case 3: Business Intelligence Prospect scans. Partner sync prep. Metrics reports. Clawdi handles the connective tissue between tools: pulls data, formats it, surfaces what needs a decision. Messy ad-hoc workflows become repeatable ops. Clawdi is built for people who want the power of OpenClaw without becoming infra engineers: • Professionals with complex workflows • Founders/operators who skipped OpenClaw setup • Security-conscious users • Small teams that want AI automation Like OpenClaw, Clawdi remembers, and learns from every interaction and gets smarter over time. Clawdi is OpenClaw for non-technical people, done right: full capability, zero configuration overhead, security that doesn't require trust. Worth a try: clawdi.ai Read on: https://lnkd.in/g3nQZCrJ

Larry Ellison just said the quiet part out loud: The AI moat isn’t the model. It’s the private data. The AI race will be won by whoever can access private data. But accessing sensitive data requires trust. And trust requires proof. That’s what Phala builds.