IdeaWell

Project ideas from Hacker News discussions.

Dots: Always-on agents

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

5 Prevalent Themes in the HN Discussion on OpenAI's Dots

1. Criticism of Cutesy/Branding Choices
Users frequently debated the appropriateness of names like "Dots," "Muse," and "Claw," with many finding them forced or mismatched to the product's purpose.

"Dots (note: not always upper-case) seems forced and impersonal, which doesn't match the vibe in the promo video." – minimaxir

2. Pricing and Accessibility Barriers
The $100+/month price point was widely seen as excluding mainstream users, conflicting with promises of mass-market appeal.

"It's not reaching mass market if it's gated behind a $100/month plan" – rolosa
"Who knows, maybe they think enterprise will pick up and run with Dots? Seems unlikely." – cartersj

3. Privacy, Security, and Control Concerns
Deep unease emerged about granting AI agents persistent access to personal data, systems, and autonomy, with references to sandboxing risks and loss of control.

"If an AI agent can't intuit how I will feel about an action it's taking on my behalf, I'm not give it access to my digital life." – jesse_dot_id
"I don't feel comfortable giving AI access to my entire computer or phone...not sure that will ever change." – dylanhouli

4. Enterprise vs. Consumer Positioning Ambiguity
Commenters split on whether Dots targets developers, general consumers, or enterprise clients, citing pricing, features, and Microsoft partnerships as evidence.

"Very, very likely targeting enterprise" – CharlieDigital (citing Microsoft Agent 365 integration)
"Muse makes sense since it targets the general normies. Dots explicitly are targeting developers instead..." – minimaxir

5. Analogies to Past Failed Assistants (Clippy etc.)
The discussion repeatedly invoked historical examples like Microsoft's Clippy as a warning about annoying, intrusive AI assistants that fail to deliver real utility.

"Does no one remember the MS Paperclip?!?" – SwabbyNat74
"I feel like this should be memed as something like the 'Jar-Jar Binks Marketing Flop'" – IAmBroom
"Clippy extends all the way here....... and not sure how they're going to avoid the comparisons" – ChrisArchitect

🚀 Project Ideas

OpenAgent Local

Summary

  • A self-hosted, open-source personal AI agent that runs entirely on the user's machine, eliminating data exfiltration and vendor lock‑in.
  • Provides a programmable harness for connecting to local LLMs (via Ollama, Llama.cpp, etc.) and tools (filesystem, email, calendar) with fine‑grained permission controls.

Details

Key Value
Target Audience Privacy‑conscious developers, power users, and enthusiasts who want full control over their AI assistant.
Core Feature Local execution of agent loops with sandboxed tool plugins (read/write toggle per tool) and model‑agnostic LLM backend.
Tech Stack Tauri (Rust + Svelte) for desktop UI, Ollama/Llama.cpp for LLM inference, WASM‑based tool plugins, SQLite for memory.
Difficulty Medium
Monetization Hobby
#### Notes
- HN users repeatedly warned about giving agents unrestricted access: “I don’t feel comfortable giving AI access to my entire computer or phone” (dylanhouli) and “Who is trusting these things with write access to their lives?” (petesergeant). OpenAgent Local addresses this by keeping all data and processing on‑device.
- The project would appeal to commenters who advocated self‑hosting: “digitaltrees: Just build your own… I think people can and should copy the full stack on top of open weights.”
- By offering a free, open‑source alternative to paid always‑on agents, it directly tackles the $100/month barrier and EU availability complaints.

ReadOnly Agent Proxy

Summary

  • A middleware proxy that sits between any AI agent (Dots, Muse, custom) and connected services, enforcing read‑only access unless explicitly approved by the user.
  • Guarantees agents can only read data (email, files, calendars) unless a temporary “write session” is started with user confirmation.

Details

Key Value
Target Audience Individuals and teams who want agent assistance but fear accidental data modification or exfiltration.
Core Feature Intercepts agent tool calls, checks policy (read‑only vs write), prompts user for consent on write operations, logs all actions for audit.
Tech Stack Node.js/Express proxy, OAuth2 token vault, WebSocket UI for consent prompts, JWT‑based session handling.
Difficulty Low
Monetization Hobby
#### Notes
- Commenters highlighted the danger of agents acting without oversight: “If an AI agent can't intuit how I will feel about an action it's taking on my behalf, I'm not give it access to my digital life” (jesse_dot_id). This proxy gives that intuition back to the user.
- The idea resonates with requests for “read‑only mode” to safely grant access: “More applications need the ability to log in as a 'read‑only' mode…” (eutropia).
- By providing a transparent consent layer, it reduces anxiety around prompt injection and uncontrolled autonomous actions.

Task‑Based Agent Lite

Summary

  • A pay‑per‑task agent service aimed at casual users who need occasional help (e.g., checking flight prices, summarizing emails) without committing to a $100/month subscription.
  • Users purchase credits or a low‑cost monthly pack; each agent action consumes a predictable amount of credits.

Details

Key Value
Target Audience Non‑technical consumers, freelancers, and small businesses who want AI assistance on a budget.
Core Feature Simple chat‑or‑voice UI where each request triggers a constrained agent workflow (e.g., “find cheap flights to Paris next weekend”) with fixed credit cost.
Tech Stack Frontend: React Native or Flutter; Backend: Python/FastAPI orchestrating LLMs (OpenAI, Anthropic, or open models) and tool adapters; Stripe for credit purchases.
Difficulty Medium
Monetization Revenue-ready: Credit packs ($5 for 50 tasks) or $5/month for 200 tasks.
#### Notes
- Many users complained about pricing: “Dots seem to me to be for the rest of general knowledge work… Requires $100/month minimum, are non‑developers paying that?” (rolosa) and “It's not reaching mass market if it's gated behind a $100/month plan”. This offers a low‑friction, affordable alternative.
- The model mirrors successful micro‑saaS (e.g., Zapier’s free tier) and addresses the desire for “cheap/free mass market like Muse” (therealdrag0).
- By tying cost to actual usage, it avoids wasteful always‑on compute complaints (“I cannot fathom how much compute will be wasted with that type of always on agentic systems” – dgellow).

AgentGuard for Enterprise

Summary

  • A governance and compliance layer that integrates with Microsoft 365, Azure AD, and Google Workspace to provide audit trails, data loss prevention, and policy enforcement for any third‑party AI agent (Dots, Muse, Grok Bot, etc.).
  • Enables CISOs to approve which agents can access which data, enforce encryption, and monitor for anomalous behavior.

Details

Key Value
Target Audience Enterprise IT, security teams, and compliance officers in finance, legal, and healthcare sectors.
Core Feature Centralized policy engine that maps agent token scopes to corporate data classifications, blocks risky actions, and generates SOC‑2‑ready logs.
Tech Stack .NET Core API, Azure Functions for policy hooks, Apache Kafka for audit streaming, React admin dashboard, SAML/OIDC integrations.
Difficulty High
Monetization Revenue-ready: Per‑seat licensing ($8/user/month) + volume discounts.
#### Notes
- HN discussion noted enterprise interest: “We’re also working with Microsoft to integrate specialist dots with their enterprise governance and security controls in Agent 365” (CharlieDigital). AgentGuard makes this turnkey for any agent.
- Concerns about data leaks were prevalent: “You look at the landscape… it's not very inspiring if you're a CISO/CIO/CTO at one of these firms.” (Reality). This tool directly addresses those fears.
- By offering a vendor‑neutral control plane, it reduces lock‑in anxiety and lets enterprises adopt agents without sacrificing governance.

Model‑Swap Agent Harness

Summary

  • An open‑source agent framework that lets users switch the underlying LLM (OpenAI GPT‑4o, Anthropic Claude 3, local Mixtral, etc.) via a simple config file, keeping all prompts, tools, and memory intact.
  • Includes benchmarking utilities to compare cost, latency, and quality across models for a given workload.

Details

Key Value
Target Audience Developers, AI enthusiasts, and power users who want to avoid vendor lock‑in and optimize for cost/performance.
Core Feature Plugin‑based LLM adapter layer (OpenAI API, Anthropic API, Ollama, llama.cpp) with unified agent state (memory, tool registry, conversation history).
Tech Stack Python (FastAPI) core, Pydantic for config, Docker‑compose for easy deployment, Prometheus/Grafana for model metrics.
Difficulty Medium
Monetization Hobby
#### Notes
- Commenters lamented lock‑in: “I really am beyond maxed out at the availability of AI's… they all are so similar now.” (ecommerceguy) and “They may develop these harness within a provider neutral company… the harness may success when combining with competitor models.” (cavoirom).
- The harness echoes the sentiment: “Tools like OpenClaw and Pi seem to remove that from the equation, letting you keep your 'history' and customization while using whatever inference provider you want.” (dpoloncsak).
- By making model switching trivial, it empowers users to chase the best price/performance and avoid being trapped by a single provider’s pricing changes.

OnDevice Agent Nano

Summary

  • A lightweight agent that runs exclusively on a smartphone or laptop using quantized LLMs (e.g., Phi‑3‑mini, TinyLlama) and leverages the device’s NPU/GPU for inference, ensuring zero data leaves the device.
  • Provides basic assistant features (reminders, note‑taking, on‑device search) with optional cloud fallback for heavy tasks.

Details

Key Value
Target Audience Privacy‑focused mobile users, travelers, and anyone wary of cloud‑based agents harvesting personal data.
Core Feature On‑device LLM inference via MLC-LLM or ggml, local tool plugins (contacts, calendar, file search), encrypted local state storage.
Tech Stack React Native (mobile) or Tauri (desktop), MLC-LLM runtime, SQLite Encrypted Extension, platform‑specific NPU delegates (Core ML, NNAPI).
Difficulty High
Monetization Hobby
#### Notes
- Privacy concerns were a recurring theme: “I don't feel comfortable giving AI access to my entire computer or phone… Seeing people give that access to agents without any sort of sandboxing blows my mind.” (dylanhouli). This removes the need for trust altogether.
- Users praised local projects: “digitaltrees: Cool. I did the same. propelcode.app… I actually greatly appreciate that I can put something cute on my sister's PC that comes from a developer that won't bundle it with malware.” (vovavili).
- The idea also addresses compute‑waste complaints by using efficient on‑device inference instead of always‑on cloud agents.

Agent UI Choice Kit

Summary

  • A set of UI themes and avatar options for any agent platform (Dots, Muse, custom) that lets users replace the default “cute” mascot with neutral, professional, or even utilitarian icons (e.g., a simple dot, a terminal glyph, or no avatar at all).
  • Includes a consent‑driven toggle to disable anthropomorphic animations and reduce perceived manipulativeness.

Details

Key Value
Target Audience Users who find cute AI avatars off‑putting or manipulative, and prefer a sober, tool‑like interface.
Core Feature CSS/theme package + JSON schema for avatar swap; works via browser extension or mobile app hook to override agent front‑end assets.
Tech Stack HTML/CSS/JS for web extensions; Kotlin/Swift for mobile overlays; Figma for theme assets.
Difficulty Low
Monetization Hobby
#### Notes
- Many comments criticized the cute branding: “Claw is a cute name. Muse is cute name. Dots… seems forced and impersonal” (minimaxir); “I feel like this should be memed as something like the 'Jar‑Jar Binks Marketing Flop'… intentionally design something to be childishly adorable…” (IAmBroom); “Panarky: A wave of nausea hit when I first saw the warm, fuzzy, cutesy, kawaii, Teletubby‑like avatar…” Offering a neutral UI directly answers that frustration.
- The kit would let users like “therealdrag0: What’s nefarious about attracting users?” choose a presentation that feels honest rather than exploitative.
- By decoupling the agent’s capability from its presentation, it encourages trust through transparency rather than cuteness‑based persuasion.

Read Later