Sabitlenmiş Tweet
English
Alex Lopez
3.2K posts
Alex Lopez
@alopez3006
Building Snipara, the continuity layer for AI software projects. I believe continuity will be as important to AI engineering as Git was to source control.
Switzerland Katılım Ocak 2012
258 Takip Edilen118 Takipçiler
I replayed real agent work on Snipara: cold start vs assisted.
Same repository.
Same task.
Very different path to useful context.
The raw diffs are here:
snipara.com/proof
I'm increasingly convinced that the bottleneck isn't context size.
It's continuity.
English
Check out Snipara on Launch Llama 🚀 tools.launchllama.co/products/snipa…
English
it does not try to restore the exact hidden LLM context or terminal process. It rebuilds the working context from explicit state.
The local workflow file is the deterministic part: goal, current phase, plan, completed phases, touched files, phase summaries, tests/checks, and next acceptance criteria. Hosted memory adds durable decisions, handoffs, and session carryover, but it is not treated as a substitute for source docs or the actual git/worktree state.
In practice, after workflow resume, the agent gets enough context to know “where am I, what phase is active, what was already verified, what files matter, and what must be done next.” Then it is expected to re-check the repo state, diffs, and tests before editing again.
We do dogfood this in real multi-phase Snipara work. A recent example was syncing the standalone snipara-companion repo to 1.4.9: after resuming, the workflow restored the active phase as “verify, commit, push, and tag”, preserved the previous phase summary, the files changed, and the checks that had already passed: type-check, lint, tests, and pack smoke. That was enough to continue without reconstructing the task manually.
The honest limitation: it is not a magic replay of the agent’s full mental state. If no phase commits or handoffs were written, resume quality drops. But when used as intended, local state + hosted memory makes context loss visible and recoverable instead of silent. For multi-phase work, that is the main reliability win.
English
@alopez3006 @X The concept is interesting, but what I'm wondering is: after a resume, how accurately does the agent actually restore the context it was in? How reliable is the combination of local state file and hosted memory in practice? Has anyone tried it in a real multi-phase environment?
English
Hi @X 👋
I'm new here and looking to connect with builders, creators, and people sharing their journey in public.
I'd love to meet:
🚀 Marketers
💻 SaaS Founders & Startup Builders
📱 Web & Mobile Developers
🎨 3D Artists
If you're building something interesting, drop a comment and let's connect.
#BuildInPublic #IndieHackers #SaaS #Startups #WebDev #MobileDev #3DArt
English
Hey founders!! Let’s #connect share your product and what you're most excited to ship next..👇 #buildinpublic
English
@sophie_launch Happy to connect — shoot me a DM if you want to chat! 💬
English
Who's building something cool in 2025? Drop your product below. Let's connect and support each other 👇 #buildinpublic
English
Hey founders/builders/tech people👋
Looking to connect with people who are into:
• SaaS
• AI Tools (tell me what you are using)
• Building in Public
• Startups
• Distribution and marketing (the most important!)
Tell me what you're working on & I'll follow you back
English
Founders, builders, 9-5ers building on the side. Lets get you some eyeballs! Drop a link to what you are working on!
I've never done this before but I have enjoyed meeting and connecting with new people this way.
I even one shotted a tool to make the image for this post called xpostmaker
English
Hey AI builders! Looking to connect with people actually shipping!
• AI Engineers
• ML Engineers
• RAG builders
• AI Agents
• LLM apps
• MCP servers
Drop what you're working on 👇
Let's discover some cool projects 🚀
#BuildInPublic #AIEngineering #LLM #RAG #AIAgents
English
Hey founders !
Looking to connect with people building in:
🈺 SaaS
👨💻 Tech
⛏️ AI tools
🔥 Web APPS
What are you working on?
English