I am a Proton Workspace Premium user. This message was drafted by Lumo to streamline the submission process.

• Feature -

I need to run the same six prompts, in the same sequence, on research markdown files daily. The output of prompt 1 becomes the input of prompt 2, continuing through to prompt 6. I want the final output without manually running each prompt and waiting for completion between steps.

• What I Tried-

Since Lumo does not support agentic or automated workflows, I explored local alternatives. I built a Python script using Ollama to host a local model (llama3.2) and execute the six-step chain programmatically.

I tested this approach thoroughly:

• Installed Ollama on Windows 11
• Configured a six-prompt sequential chain
• Tested with placeholder text and actual research files

• Verified the automation logic works correctly

• What Happened-

The workflow functioned technically, but the local model produced unacceptable levels of hallucination. Even with careful prompt engineering, Llama 3.2 (3B parameters) fabricated information throughout the chain. Errors compounded at each step, rendering the final output unreliable.

I attempted mitigation by switching to larger local models, but the hardware constraints of a 16GB laptop prevented me from running models capable of handling multi-step reasoning without hallucination.

• The Privacy Constraint -

Because of the nature of the material I process with Lumo, I cannot reliably trust public cloud providers like OpenAI, Anthropic, or Google. This is not a technical limitation but a fundamental requirement tied to the purpose of zero-knowledge encryption.

I chose Proton and Lumo specifically because data privacy is the core value proposition. If users must abandon that promise to obtain reliable AI performance, the tradeoff defeats the purpose of choosing a privacy-first service in the first place.

• The Request -

I need a way to run automated multi-prompt workflows with reliability comparable to Lumo's manual chat interface. The implementation details are not my concern; I do not care if this runs locally, on Proton servers, or through hybrid infrastructure. What matters is:

• The ability to chain prompts sequentially with state passing
• Output quality that does not degrade through the chain

• Privacy guarantees that align with Proton's zero-access encryption model

• Why This Matters to Proton -

Users who depend on Proton for privacy-sensitive work need the same automation capabilities available in non-private AI services. Otherwise, I face a binary choice: compromise privacy for functionality, or sacrifice efficiency for security. Neither is acceptable.

Proton's mission is to build an internet where privacy and security are the default. That mission extends to AI capabilities. Automated workflows should not require users to abandon the very privacy protections they signed up for.

Thank you for considering this request. I appreciate the work Proton has done to differentiate itself in an AI landscape dominated by data-harvesting providers. I believe this feature would strengthen that positioning and serve the needs of privacy-conscious professionals.