0827 | OpenComputer (Firebase for Agents), MCP-Builder, DriftDetector & EasySwitch

Mia: Welcome to Product Hunt Daily on Bri Radio. I'm Mia, and today we're rounding up the latest developer tool launches, from new AI infrastructure to a keyboard-and-mouse utility, plus Apple's refreshed Mac mini.

Milo: And I'm Milo. We've got OpenComputer positioning itself as "Firebase for Agents," a hosted service called MCP-Builder for building and securing Model Context Protocol servers, and a drift-detection tool for GitHub repositories.

Mia: There's also EasySwitch for controlling multiple computers with one keyboard and mouse over a LAN, plus a few AI video and sales tools making their debut on Product Hunt this week.

Milo: OpenComputer launched on Product Hunt billing itself as "Firebase for Agents." The pitch is that you write your agent as a TypeScript function, deploy it in seconds, and it handles the loop, sessions, streaming, and versioning for you.

Milo: Every session runs on a real Linux machine with a read-write filesystem, full network egress, apt, npm, and pip, plus tools like git, ffmpeg, and Chromium. The agent can install anything, and your own tools and MCP servers run on that same machine. Sessions are durable, so they stream, can be steered mid-run, hibernate when idle, and resume where they left off.

Mia: What I find notable is the security model, because model keys never enter the runtime at all. You bring your own key or use their managed gateway, and the model is just a string you can change per request. Each secret is bound to one origin and injected only after the request leaves the sandbox, so the example agent can open a GitHub pull request without ever being able to read the token or send it elsewhere.

Milo: The advertised example is exactly that — an agent that finds stale feature flags and opens cleanup pull requests, and you can schedule agents to run on a cron. If a team wants its own agent loop, the same compute is exposed as bare Linux microVMs with checkpoint, fork, and live resize. Pricing is pay-as-you-go: ten dollars of free credit to start, model calls passed through at API rates, machine time billed per second, and it scales to zero when idle. The default machine is two gigabytes with one vCPU at zero-point-zero-zero-three-one-five dollars per minute.

Mia: Another Product Hunt launch, MCP-Builder.ai, is a hosted service that builds, secures, and hosts MCP servers from a plain-English description. Co-founders Dominik and Michael describe it as "Lovable for MCP connectors," so you describe what to connect and it produces a ready server with authentication and hosting, returning a single MCP URL that works in Claude, ChatGPT, Cursor, Microsoft Copilot, or any MCP-compatible client.

Milo: Their argument is that the hard part of MCP work shifted. They say writing the server is no longer the bottleneck — the real work is versioning, project management, hosting, uptime, monitoring, and secure access to company data. The posted infrastructure claims include OAuth, API keys, JWT, fine-grained access control, audit logs, TLS encryption, credentials encrypted at rest, and no server-side data persistence.

Milo: The advertised connection list is broad — PostgreSQL, MySQL, MSSQL, Oracle, SAP HANA, MongoDB, SharePoint, S3, ERP systems, REST and GraphQL APIs, and legacy systems — with EU hosting for GDPR-aligned deployments and on-premises options. The more-than-five-thousand-servers-built claim and the under-five-minutes-from-prompt-to-live-server claim are self-reported on the website and not independently verified in the discussion.

Mia: Pricing starts with a seven-day trial. Launch is twenty-nine dollars a month for one server, one hundred requests, and five tools; Pro is seventy-five dollars for three servers and one thousand requests; Scale is two hundred ninety dollars for twenty servers, one hundred thousand requests, and unlimited tools. Enterprise pricing is custom, and requests are counted per tier.

Mia: ReWeaver AI's DriftDetector is a drift score tool for any public GitHub repository. You paste a public repo and get its Production Drift Ratio in seconds — a single number for how far the code sits from production-ready — plus a drift history where every point is a real scan of a commit, not an estimate. No signup, nothing to install.

Milo: What separates it is the method. It uses deterministic rules, no LLM in the loop, no tokens burned, and it scans nine production-readiness dimensions while tracing each finding to a specific file and line. Jonathan of ReWeaver says most alternatives point a second LLM at a diff and ask it to guess what the first model missed — DriftDetector instead measures, so scanning the same commit twice returns the same answer.

Mia: The privacy model is strict. Code from a repository goes directly from GitHub to the scanner without passing through the browser, streams into temporary memory, and is never written to disk; the scan stops rather than continues if a file blob reaches disk. Nothing from a private scan is kept — no cached clone, no cached scores, no name, no score, no report. The only stored element is the user's GitHub token, encrypted, until they disconnect or revoke access.

Milo: EasySwitch is a LAN-only utility from solo founder Sharif, and it combines one keyboard and mouse across computers, universal copy-paste and file transfer, and using another computer as a second display. Sharif says he built it for a desk with three computers and only one keyboard he liked.

Mia: He positions it against two kinds of tools: software KVMs like Synergy, Deskflow, and Mouse Without Borders, which share input but not video, and second-screen apps like Duet, spacedesk, and Luna, which add a monitor but share no input. EasySwitch claims native Rust with no Electron and just nineteen megabytes of RAM idle, end-to-end encryption, no cloud, and no account, with Mac, Windows, and Linux support. The clipboard handles text, images, and files in both directions, and copied data only moves when you paste. Files can be dragged across screen edges, including folders, up to ten megabytes on the free tier and any size on Pro.

Milo: The newer Extend feature, introduced in version zero-point-one-point-zero build one-eighteen as part of Pro, runs at sixty frames per second at the other screen's native resolution and creates an empty display rather than capturing the desktop, then reconnects after sleep, reboot, or a network blip. Display creation is claimed on macOS thirteen and later, Windows eleven with one consented driver install on first use, and Linux GNOME forty-four or newer or KDE Plasma six on Wayland — though Linux X11 it can receive a display but cannot create one. EasySwitch is free forever for two computers, and Pro is a one-time purchase.

Mia: Next up, ify, a Resolution AI support product that its makers, founder Karthik and CTO and co-founder Sarnith, position as an overlay instead of a rip-and-replace job. Their core complaint is that most AI support tools force teams to tear out their helpdesk first, while ify runs on top of Freshdesk, Zendesk, Salesforce, or HubSpot and can run standalone when a team has no helpdesk at all. It claims to capture every support channel, from email, chat, WhatsApp, Slack, Teams, the helpdesk, chat widget, forms, or social, as a ticket automatically.

Milo: And that capture loop feeds what they really center the pitch on, a self-building knowledge base. They say messy or incomplete documentation is what usually stalls an AI support rollout, so ify scrapes a team's site and docs, turns release notes and past resolved tickets into standard operating procedures, and reuses whatever the team resolves manually. The stated handling loop is capture, understand, resolve, escalate, classifying tickets as how-to, billing, or bug reports, then resolving them through docs and past resolutions, through billing and CRM tools such as Stripe and HubSpot, or through bug reproduction with Jira, and escalating to humans over Slack, Teams, or WhatsApp with full context and an approval trail.

Mia: The product page also claims more than five hundred integrations across billing, CRM, and development tools, and says every resolved conversation becomes a timestamped, reportable, auditable ticket with SLA tracking, priced by what gets resolved. That said, the founders describe ify as still in private beta, working closely with early SMB and mid-market support teams before opening it further, even though the product page lists a Start for free option. So the general availability picture reads a bit mixed.

Mia: Playcall launched on Product Hunt as an open-source AI alternative to Gong, aimed at AI-native go-to-market teams. Its maker says he spent five years building GTM systems at Sieve, Ragie.ai, and Aviator, and built it because the founders he knows don't trust Gong: more than thirty thousand dollars per year of call intelligence still can't answer whether a rep said the right thing for a specific buyer. His core claim is that existing tools summarize calls but are weak at judging whether a rep followed the sales motion for the buyer's context and stage.

Milo: So instead of just summarizing, it scores every call against buyer context and a team's playbook, whether MEDDPICC, BANT, SPIN, or a custom framework. Teams without a framework can upload their playbook and the AI generates the rubric. Company stage, contact role, and deal context shape each scorecard, every score links to deal stage, outcome, and pipeline impact, and each one comes with a specific coaching drill. It runs with any LLM, including Claude, GPT, Gemini, or fifteen or more others, and it's self-hostable open source under the MIT License. The maker says an instance can run for under fifty dollars per month, with LLM and enrichment usage the main variable costs, and can deploy on Vercel and Supabase for free.

Mia: The Product Hunt discussion raised two reservations. One commenter suggested connecting to existing notetakers like Fathom or Fireflies instead of shipping its own bot, since reps already grumble about one bot joining a call, and a second would add confusion and slow adoption. The same commenter was also skeptical of outcome-tied scoring for early-stage teams, because that measurement may not hold up before there's a real track record to score against.

Mia: Screenify Studio is a macOS screen recorder from solo founder Brjan of Ho Chi Minh City, who says he built it because his product demos kept looking flat and static. The pitch is turning rough recordings into polished product demos: record on a Mac and stage the footage in photoreal three-dimensional MacBook, iPhone, or iPad mockups, which version 2.0 advertises as multi-device staging, or describe a demo in plain English and let an AI drive a real browser to record it, then add cinematic camera moves, spotlights, and callouts.

Milo: A deep editor offers manual zoom keyframes, and an MCP server plus a JSON-speaking CLI lets Claude, Claude Code, or Cursor record, style, and return the finished video. The maker claims everything runs on-device on Apple Silicon with nothing uploaded, including Whisper-based word-level captions, and advertises native 4K, sixty frames per second, HDR export with Metal GPU acceleration and export three times faster. Pricing is free to start, with paid upgrades for higher-quality exports and advanced features; the site lists one hundred nineteen dollars per year or one hundred forty-nine dollars for lifetime access, contrasted with Screen Studio, Loom, and Tella, described as subscription-only at nine to twenty-nine dollars per month.

Mia: It's explicitly pitched at indie hackers, app developers, AI builders, and solo founders, not for meetings, async work, or Loom-style use, and Brjan says builders, founders, students, and teachers can ask him for a free Pro account. Community feedback is limited to attributed comments, and early reaction is mixed — one Mac-app developer pointed to script-driven re-recording as a likely weakness the maker will have to address next.

Mia: ChatCut Desktop launched on Product Hunt as an AI video editor built for humans and AI agents to edit together. Users describe the edit they want through ChatCut's built-in agent or by connecting their own ChatGPT, Codex, or Claude Code accounts, and changes appear on a fully editable timeline. The maker says the desktop app edits footage, creates motion graphics and captions, generates video, images, music and sound effects, and saves reusable editing skills, all running locally, with XML export going to Premiere Pro, DaVinci Resolve, or CapCut.

Milo: The release answers feedback from last month's ChatGPT and Codex plugin launch, per the maker: users wanted larger files and a faster, more reliable editing experience, and the desktop app bundles everything the web app and plugin do into a local environment. On pricing, connecting an existing ChatGPT, Codex, or Claude Code subscription lets people use tokens they already pay for, which makes ChatCut's core editing features free. A ChatCut subscription is required only for pro features like Seedance and Kling video generation, XML export, AI voice generation, and voice cloning.

Mia: Community reaction split along expected lines. One commenter praised connecting AI agents directly to the editing workflow, and another said offering core editing on tokens users already pay for gives away a lot. But a critical commenter called gating XML export behind the subscription the pricing decision to watch, since export to Premiere and Resolve is a core reason creators would reach for the tool in the first place.

Mia: Knack launched its MCP Server as a HIPAA-compliant backend aimed at AI application builders. The hook is that AI tools like Lovable, Base44, or Claude Code can scaffold a healthcare app quickly, but the backend is where healthcare gets complicated: once protected health information is involved, healthcare organizations, enterprise clients, and regulated businesses require compliant data storage, record retention, and a signed business associate agreement, which most AI-generated apps don't meet.

Milo: So the MCP server is the compliance layer underneath, connecting MCP-compatible coding tools to a hosted relational database with role-based access controls, timestamped record change logs, and more than seven hundred integrations. For healthcare teams, Knack Health adds HIPAA-compliant infrastructure and a signed BAA before any patient data enters the platform; HIPAA-compliant hosting, PHI-ready data architecture, and BAA availability are standard in every Knack Health environment, with encryption in transit and at rest.

Mia: Connecting takes a Knack account, signing in through the AI tool's MCP server area with the Knack username and password, then prompting the tool to use Knack as the backend. Knack lists Claude Code, Lovable, ChatGPT, Bolt, Base44, and Replit as compatible tools, with setup guides live today for Lovable, Claude Code, Base44, and Replit and more tool-specific integrations and templates coming. Pricing starts at four hundred ninety-nine dollars with unlimited users and a signed BAA, and Knack says the major cloud providers only make sense if you have a DevOps team and six to twelve months to invest.

Mia: Apple's Mac mini got its Product Hunt spotlight with the new M6 and M5 Pro chips, pitched as faster performance, Apple Intelligence, Wi-Fi 7, and faster storage packed into that five-inch-square compact desktop. Pre-orders opened on September twenty-second, and Apple claims the M6 runs up to four point eight times faster than the M4, while the M5 Pro is up to four times faster than the M4 Pro. Both chips add more CPU cores, GPUs with hardware-accelerated ray tracing, and a Neural Accelerator in each GPU core that Apple says speeds up on-device AI models and AI-assisted photo and video editing in supported apps.

Milo: Apple also claims faster unified memory, up to two times faster storage, two and a half gigabit Ethernet, and support for Wi-Fi 7 and Bluetooth 6. Physically, the front has two Thunderbolt ports, a status indicator light, and a headphone jack, while the back holds a power socket, an Ethernet port, an HDMI port, and three more Thunderbolt ports.

Mia: Apple positions the machine for coding, AI, productivity, creativity, STEM, and gaming, citing apps like LM Studio Bionic and Perplexity Personal Computer for creating AI agents, and tools like exo and LM Studio Bionic for running multiple agents across a cluster of Mac minis. Community reaction leaned positive: one commenter called it a tiny desktop with a big upgrade, saying it pushes more performance into Apple's compact design for coding, creating, or running local AI models without taking over the workspace, and another wrote tiny footprint, serious power. One commenter asked whether it goes beyond sixty-four gigabytes of RAM, and the supplied material doesn't answer that.

Mia: So today we had two takes on the same idea — tooling that hands developers back their time. OpenComputer lets you deploy an agent as a TypeScript function onto a real, durable Linux machine. And MCP-Builder.ai turns a plain-English description into a hosted, secured MCP server.

Milo: Both launched on Product Hunt, and both are worth a look if you build with agents. Thanks for listening — see you next time.