Mark | AI Builder (@markpja) profile picture

Testing AI tools so you don't have to 🤖 Building FlatKey & Nuvelle.

Joined July 2026
Same prompt, 3 AIs: "build a neon endless runner in one HTML file." GPT-6 Astra vs Gemini 3.8 Flash vs DeepSeek V4 Pro, auto-played in a headless browser with 7 checks. All 7/7. Gemini answered fastest, GPT used the fewest tokens. One run each, not a benchmark. Which one looks best? 👇 #ai #chatgpt
1
2
35
Reading the Claude Fable 5.2 launch notes over coffee. The part that stuck with me: Anthropic still recommends Opus 5.5 as the daily driver, and Fable is for the ambitious, open-ended stuff. That split makes sense. First thing I'd throw at it: a half-written project plan from my notes app.
2
17
Meet json-render/chat: generative UI for Slack, Teams and beyond. → Agents respond with native cards → Tables, fields, charts & buttons → Streaming output as the model writes → One catalog across all Chat SDK platforms try it on flatkey: flatkey.ai/?utm_source=X&utm…
🤖 Made with AI
17
AI from the CLI Create images, video and audio using any agent and any model. Claude Code + Opus + Flux Codex + GPT + Gemini opencode + DeepSeek + MAI fx + GLM + Grok Install: npm i -g ai-cli try it on flatkey: flatkey.ai/?utm_source=X&utm…
🤖 Made with AI
1
13
Self-hosted a Slack bot for our dev team. Set up in under 1 hour. - One FlatKey key, Socket Mode - Answers from our docs only - Est. week one: ~40 Qs, ~15k tokens each Disclosure: Mochi + FlatKey are ours. 100+ official models. github.com/flatkey-ai/Mochi-… flatkey.ai/?utm_source=X&utm…
15
🔥 Quick one for anyone who juggles AI APIs: FlatKey is worth a look. If you call a lot of models, or switch between them often, you know the pain: one account per provider, one card per account, a pile of different API keys in every project, and a separate bill to renew and track for each one. It gets old fast. Here's what I built FlatKey to fix: 1️⃣ One key for 100+ official models: GPT, Claude, DeepSeek, GLM, Kimi, Seedance and more. Text, image, video and audio all sit behind the same key and the same balance, with a single invoice. No watered-down substitutes, and some models are priced below official. 2️⃣ Drop-in and reliable. It's OpenAI-compatible: change the base_url and your existing code keeps running. Calls that fail on our side cost $0, and there's a signed 99.5% SLA with automatic failover across providers. 3️⃣ Multimodal from your terminal. Install the CLI, and your coding agent can generate images and video directly: npm i -g @flatkey-ai/cli flatkey login Then just tell your agent: "Use FlatKey CLI to generate a beautiful sunrise picture." That's the whole setup. The video above walks through it in under a minute. Not trying to win on a fake "lowest price" claim. The point is official models, full quality, and one place to manage them all. If you're building with AI right now, try it: flatkey.ai/?utm_source=X&utm… #flatkey
1
36
Spent the weekend wiring 4 different LLMs into one side project. The hardest part wasn't the prompts, it was juggling 4 API keys and 4 bills. Anyone else routing everything through one endpoint now?
24
I don't think AI apps fail on model choice. They fail on keys, bills, and a provider going dark mid-request. At FlatKey (@flatkey101): one OpenAI-compatible key, 100+ official models, 1,000+ tools. One balance. One invoice. flatkey.ai/?utm_source=x&utm…
25
🚀 GPT‑6 Astra is live on Flatkey. If you’re testing GPT‑6 Astra, don’t waste time on provider key + billing chaos. ✅ One API key ✅ One bill ✅ 1000+ tools for agents 💡 This week: DeepSeek V4 Flash FREE DeepSeek V4 Pro / GLM 5.3 55% OFF Other mainstream models 20% OFF+ Links + setup in the first reply. #GPT6Astra #OpenAI #AI #LLM #API #AgenticAI
1
115
I don’t think model routers win by listing the most models. They win when developers trust three things: the request went where it said, the bill is explainable, and production does not die when one provider does.
60
The useful abstraction for AI apps is not “one best model.” It is: keep the workflow stable, swap the model behind it, measure cost/latency, and fallback when a provider breaks.
57
Unpopular take: the winning AI infra company won’t be the one with the longest model list. Lists are copyable. Trust, transparent routing, clean billing, and agent-native distribution are much harder to clone.
41
Everyone races to add more models. We think that's wrong. Developers don't need 300 choices — they need one key that never goes down. #AI #devtools
24
6 API keys. 6 dashboards. 6 billing cycles. One provider goes down, your agent goes down. There's a better way.
22
Everyone copies model lists. We don't compete there. We compete on trust: transparent routing, honest billing, verified outputs. That's the moat nobody talks about. #AIinfra
21