HN Showcase
Back to browse
LM Canvas – current chat interfaces suck, so I built a canvas for LLMs

LM Canvas – current chat interfaces suck, so I built a canvas for LLMs

by max-lee-dev·Mar 5, 2026·2 points·1 comment
Visit ProjectView on HN

AI Analysis

●MidBig BrainBold BetShip It

Multi-model canvas is clever, but 300+ model support and branching don't solve core UX friction vs. Continue, Claude projects.

Strengths
  • •Parallel execution and branching (memory management per branch) address real pain points in exploratory LLM workflows.
  • •300+ model support with zero lock-in is genuinely useful for researchers and comparison-minded users.
  • •Spatial canvas UX sidesteps linear conversation degradation—fresh design language for LLM interaction.
Weaknesses
  • •Twitter link leads to post, not a working product page—no live demo, no sign-up, no evidence of shipping beyond hype.
  • •Branching + multi-model conversation management is conceptually appealing but unproven; ChatGPT Projects, Claude artifacts, and Continue already offer partial solutions.
Category
AI/ML
Target Audience

Power users of LLMs, researchers comparing models, developers wanting fine-grained conversation control.

Similar To

Continue.dev · Claude Projects · ChatGPT Canvas

Post Description

hi guys! i’m 19, been coding since I was 8, and during high school worked as a SWE at Interaction/Poke (100M) and Founding Engineer at build.ai (100M). i’m building out lmcanvas.ai during my gap year after the demo hit 2.1M views, 15K likes on X.

i hated current chat interfaces for a long time:

1. Context switching hell — the moment you want to explore a different angle you either muddy the current thread or start over and lose everything you built up.

2. Memory degradation — LLMs get worse as conversations grow. Some queries need no memory, some need temporary context, some need everything. Branching solves this cleanly.

3. Parallel execution — why wait for one response when you can run multiple simultaneously? Every chat app makes you wait. That's absurd.

4. Model lock-in — OpenAI won't show you Claude's answer. Anthropic won't show you GPT's. LM Canvas connects to 300+ models and has no reason to hide any of them from you.

so i built a spatial canvas for working with LLMs — branch conversations, run prompts in parallel, compare models side by side. you can also import your ChatGPT conversations and they convert into branchable nodes on the canvas.

please give me the most spicy / raw feedback you have to offer. i’ll be in the comments for all day / night :)

Similar Projects

Developer Tools●●Solid

NadirClaw – Open-source LLM router with 10ms classification

Smart LLM routing cuts costs, but competing against established OpenRouter and vLLM ecosystems.

Solve My ProblemBig Brain
amirdor
103mo ago
AI/ML●●Solid

LLM-DAG-UI – A branching conversation interface for Claude

Drag-to-branch is the clear selling point: you can fork any message and pursue alternate prompts while that branch only inherits direct ancestors. The BYOK approach (API key stored in your browser + local Express proxy) keeps it lightweight and safe for a demo, but it's clearly a concept build tied to Claude rather than a multi-model, production-ready tool.

Niche GemRabbit Hole
tenahu
104mo ago