Launching today
Bob's CLI
A local-first AI coding CLI that adapts to you
150 followers
A local-first AI coding CLI that adapts to you
150 followers
Bob's CLI runs on your own hardware with zero API costs, zero data leaving your machine. Bob lives in your terminal, sees your actual files, and writes code only with your explicit approval. What makes it different: auto-detect local AI models, behavioral DNA profiling that adapts to how YOU work, autonomous code review + auto-fix, conversation forking, deep dives, and SovereignLink — remote execution from any device while your code stays home. Free to start. Sovereign by design.
BobCA
Hey Everyone 👋 Kemone here, founder of Bob's Workshop.
I built Bob's CLI because I was tired of a lie this industry keeps selling: that you need to quit your job, raise capital, and go all-in to build something real. That's gatekeeping dressed up as hustle culture. The truth? You can keep your day job, pay your bills, and still ship production-grade software — if your tools respect your time.
The Reality for Most Builders:
You're a developer with a 9-5. You have a side project — maybe a SaaS, maybe an app, maybe something that could change your life. But you only get 45 minutes on the train. An hour before bed. A lunch break at your desk. And every time you sit down, you waste half that time just remembering where you left off.
What Bob's CLI Actually Solves:
Bob is a senior AI engineering partner that lives permanently in your terminal. He remembers your entire project architecture. He remembers your last conversation. He remembers YOUR patterns — how you think, how you decide, how you build. So when you open that terminal at 6am before work, there's zero ramp-up time. You're immediately productive.
But here's where it gets real — SovereignLink. Start bob serve on your desktop before you leave the house. Now your home machine is a personal AI cloud. On the bus? Send commands from your phone through the web app. At a coffee shop with a Chromebook? bob remote chat "add the payment webhook". Your desktop at home receives it, reads your actual files, generates the fix, writes it to disk. You come home and the feature is waiting for you.
No dropout required. No VC required. No risk required.
Just a developer with a vision, a terminal, and a tool that refuses to waste their time.
What Makes This Different From Every Other AI Tool:
Runs on YOUR hardware — zero API costs on the free tier
Your code NEVER leaves your machine unless you choose
Auto-detects your local model — type bob chat "hello" and it just works
Behavioral DNA profiling — Bob adapts to your style over time
Conversation forking — explore ideas without losing your thread
Autonomous code review — Bob finds bugs while you sleep
One command to push: bob push "shipped it" — stage, commit, push, done
Who This Is For:
The developer who drives for DoorDash but has an app idea. The teacher coding between classes. The parent building after bedtime. The full-time employee who's also a full-time dreamer. You don't need to bet your livelihood to build something incredible. You just need a tool that multiplies the limited time you have.
Install it:
Bash
Your AI. Your hardware. Your schedule. Your future.
I'm in the comments all day — tell me about the project you're building in the margins of your life. I guarantee Bob can help you ship it faster. 🌱
Local-first for a coding CLI is a genuinely interesting architectural choice. Keeping code context local avoids round-trip latency and the data-residency concern that blocks enterprise adoption of cloud-based tools. What's powering the 'adapts to you' part? Is it RAG over the local codebase, fine-tuning on usage patterns, or closer to behavioral prompting based on past interactions?
BobCA
@anand_thakkar1 Great breakdown of the tradeoffs you get the space. To answer your question: it's all three, but layered.
Project context uses RAG over the codebase that handles the "what are you building" awareness. Combined with summarization and dynamic file mapping. But the personalization layer is a completely separate proprietary architecture. Our Frank Engine (built alongside NVIDIA and Google Cloud) maintains a dual-structure memory system short-term interaction patterns and long-term behavioral DNA that profiles how you think, how you make decisions, and what approaches have failed for you before. It's not retrieval. It's not fine-tuning. It's a purpose-built personality engine that self learns your engineering philosophy over time so Bob's responses aren't just contextually accurate they're you-accurate.
Local-first with auto-detected local models is a great angle, especially the no-API-bill part. Curious about the day-to-day feel: on a normal laptop with a local model, is it fast enough to keep in the loop while coding, or more of a background reviewer?
BobCA
@ianhxu We've tested it on the NVIDIA RTX 5070 Ti (via Acer Predator) and the NVIDIA RTX 4050 (via HP Victus). On the RTX 5070 Ti, responses come back in 2-4 seconds with a 7B model — fast enough to stay in your flow while coding. The 4050 is slightly slower (4-8 seconds) but still very usable for iterative chat. For Apple Silicon machines like the MacBook Neo (A18 Pro, 8GB unified), a model like Gemma 4 E4B fits comfortably in memory. Based on the hardware specs and Apple's Neural Engine acceleration, we'd expect similar response times to the 4050 solidly in the keep in the loop while coding range.
Nice angle. For a local-first coding CLI, the thing I’d want to see is how it handles the boring failure cases: failed tests, half-applied edits, and leaving a diff that is easy to review or roll back.
BobCA
@kevinzrzgg Really appreciate this question these are exactly the failure cases we obsess over. Today, every file edit goes through an approval gate (nothing writes without your explicit yes), and every approved write auto-backs up the original to .bob-backups/ with a timestamp. Our auto-fix agent also runs programmatic validation before writing if the output looks corrupted, it's rejected and the file is never touched.
Honest gap: we don't have a dedicated bob rollback command yet, thanks to your questions that's now listed as something mandatory to be implemented starting today. Until then, recovery is .bob-backups/ or git checkout. Expect a proper rollback flow in the coming days. (we release updates EVERY day... 😊)
@kemone_phillips @amanda_phillips6 Local-first is a strong direction for coding tools. The more an AI assistant learns from your workflow, the more important it becomes that developers understand what context is being used and can keep control of the environment.
BobCA
@amanda_phillips6 @alpertayfurr 100% agree sovereignty over your own context is non-negotiable for us at Seedling. That's why everything Bob learns about you stays on your machine by default. Your code, your behavioral profile, your conversation history all local in ~/.bob/.
Nothing leaves unless you explicitly choose to sync. Full transparency, full control, always.
Give it a spin: pnpm add -g @bobsworkshop/cli
#DreamBuildGrowTogether🌱
Voquill
This looks promising, especially for builders who prefer keeping everything local. I'd love to know what the recommended hardware setup is for running this effectively. Congrats on the launch!
BobCA
@henry_habib Hey Henry! Thanks for the kind words 🙏 Entry level (8GB RAM, integrated GPU): MacBook Neo, MacBook Air, or any budget laptop runs Gemma 4 E4B or Mistral 7B. Responses in 2–5 seconds. Totally usable for daily coding. Sweet spot (16–32GB RAM, RTX 5060–5070 Ti): This is where most developers land. Runs Gemma 4 12B or DeepSeek-R1 7B. Responses in 1–3 seconds. Feels like pair programming in real-time. Powerhouse (32–64GB RAM, RTX 5080–5090): Runs 70B parameter models. Sub-2-second responses. Can also serve your entire team via SovereignLink...one machine, multiple developers connecting remotely. Happy to help you figure out the ideal setup for your specific hardware if you want to drop your specs!
Stripo.email
Congrats on the launch! The idea of an AI partner that remembers project context and works around a developer's limited time is really compelling.
BobCA
@alina_tyslenok_ Thank you! That's exactly the problem we built for most of us aren't coding 8 hours straight, we're grabbing 45 minutes between meetings or an hour after the kids are asleep. Bob remembers everything between sessions so you never waste those precious minutes re-explaining context. Pick up exactly where you left off, from any device.