AI Chat

The AI chat is the primary way to interact with PlasmidStudio. It can:

- Design plasmids from natural language descriptions - Modify existing constructs (add/remove/replace features, tags, regulatory elements) - Answer questions about features, compatibility, expression systems, and cloning methods - Generate protocols with materials, steps, primers, and verification strategies - Suggest cloning strategies based on your construct's characteristics - Optimize codons for target organisms

Starting from scratch: - "Design a GFP expression vector with CMV promoter and kanamycin resistance" - "Create a Gateway entry clone for human TNF-alpha" - "Build a dual-promoter construct with GFP and mCherry"

Modifying a construct: - "Add a 6xHis tag to the C-terminus of GFP" - "Replace the ampicillin resistance with kanamycin" - "Add a T2A self-cleaving peptide between GFP and mCherry"

Asking questions: - "What expression system should I use for this construct?" - "Are there any restriction sites I could use for subcloning?" - "Explain the Design Health warnings"

Messages in the chat are visually distinguished:

- Your messages — teal left accent border, right-aligned - AI responses — dark background with model name label (e.g., "Claude") - System messages — compact muted italic styling - Error messages — amber left border

When the AI generates a construct, an "Apply to Map" button appears. Click it to load the construct into the viewer.

PlasmidStudio supports multiple AI models: - Claude Sonnet 4 (default) - Claude 3.5 Haiku (faster, good for simple tasks) - GPT-4o - GPT-4o Mini

Switch models from the dropdown in the chat topbar. You can also bring your own API key (BYO) in Settings — keys are encrypted server-side with AES-256-GCM and never exposed in the browser.

AI designs should always be verified. The AI can make mistakes — wrong enzyme sites, incorrect orientations, or hallucinated properties. Design Health checks catch many of these, but not all.

Rate limits apply: 30 chat requests per hour, 6 analysis requests per 10 minutes.

The AI works with public knowledge. It won't know about proprietary sequences or unpublished modifications unless you provide them in the conversation.