Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/wtyler2505/ProtoPulse/llms.txt

Use this file to discover all available pages before exploring further.

ProtoPulse’s AI assistant is not a chatbot that gives advice — it is a design agent that executes actions directly on your project. When you type “design a 3.3V regulated supply from USB-C,” the AI places architecture nodes, draws signal edges, populates the BOM with real part numbers, runs a design rule check, and streams every step back to you in real time. All 82 actions are typed, validated, and undoable.

​
What makes this different

Most AI tools in EDA software are documentation search wrapped in a chat interface. ProtoPulse’s AI has 82 tool actions spread across 11 modules. Every action maps to a real operation — writing to the database, triggering a client-side state update, or generating a file download. The AI does not simulate actions; it runs them. Key properties of the action system:
  • Typed and validated — every tool call goes through a Zod schema before execution. Malformed parameters are rejected with a clear error rather than silently ignored.
  • Undoable — every AI action that modifies design state is logged to the audit trail. Use the undo tool action (or Ctrl+Z) to reverse any step.
  • Destructive actions require confirmation — operations that delete data (removing all nodes, clearing validation issues, deleting a BOM item) trigger a confirmation dialog before proceeding. The AI cannot clear your design without your explicit approval.
  • Audit logged — every AI action writes a record to the ai_actions table. The Audit Trail view shows the full history of what the AI did, when, and with what parameters.

​
Supported AI models

ProtoPulse supports two AI providers. You can switch between them at any time without losing your conversation history.
ProviderModelsStrengths
Anthropic Claudeclaude-3-5-sonnet, claude-3-opus, and othersDeep reasoning, multi-step circuit design, long context
Google Geminigemini-pro, gemini-flash, and othersSpeed, cost-efficiency, vision capabilities
Both providers stream responses via Server-Sent Events (SSE), so you see text and tool-call results appear progressively rather than waiting for a complete response. The active provider and model are shown in the status indicator at the top of the chat panel.

​
How streaming works

When you send a message, the server opens an SSE stream to the selected AI provider. The stream emits a sequence of typed events:
  • text — a chunk of the AI’s prose response
  • tool_call — the AI has decided to invoke a tool (name + parameters shown)
  • tool_result — the tool executed; success or failure reported
  • done — the full response is complete; all actions are summarized
The chat panel renders each event as it arrives. Tool calls appear as inline action cards showing what the AI did and whether it succeeded. If a tool fails, the AI receives the error and can adjust its approach.

​
Configuring your AI settings

Click the gear icon in the chat panel header to open AI Settings.
1

Choose a provider

Select Anthropic or Google from the AI Provider dropdown. Each has distinct model options and requires its own API key.
2

Enter your API key

Paste your API key into the API Key field. ProtoPulse stores it encrypted on the server (AES-256-GCM) and never exposes it in the browser or logs.
3

Select a model

Choose from the model dropdown or type a model name directly. If you are unsure, the default model for each provider is a good starting point.
4

Tune optional parameters

Adjust Temperature (0–2, default 0.7) to control how creative vs. precise the responses are. Adjust Max Tokens (256–16,384) to control response length. Add a Custom System Prompt to inject standing instructions — for example, “Always prefer LCSC stock” or “Use STM32 over ESP32 when performance matters.”
Your API key is encrypted with AES-256-GCM before it is stored in the database. The encryption key is set by the API_KEY_ENCRYPTION_KEY environment variable on the server. Your key is never logged, never returned to the client, and never included in error messages.

​
The 11 tool modules

Every AI action belongs to one of 11 modules. The table below summarizes each module’s scope.
ModuleScope
ArchitectureAdd, remove, and update block diagram nodes and edges; generate complete architectures; auto-layout; manage hierarchical sheets; assign net names; set pin maps
CircuitCreate schematic designs; place and remove component instances; draw and remove nets; place power symbols, no-connects, and net labels; run ERC; place breadboard wires; draw PCB traces; auto-route
BOMAdd, remove, and update BOM items; export CSV; look up pricing; suggest alternatives; optimize BOM; parametric search; attach datasheets; compare components
ValidationRun DRC; add and clear validation issues; power budget analysis; voltage domain check; auto-fix common issues; DFM check; thermal analysis; generate test plan
ExportGenerate KiCad, Eagle, SPICE, BOM CSV, Gerber, drill files, pick-and-place, netlist (KiCad + CSV), design report, and Fritzing project downloads
ComponentCreate, modify, and delete component parts; fork library components; run DRC on a specific component
NavigationSwitch between the 30 application views; jump to a specific schematic sheet
ProjectRename the project; update the description; save design decisions with rationale; add annotations; undo/redo
SimulationGenerate SPICE netlists; run AC/frequency analysis; retrieve simulation results
AnalysisAnalyze uploaded images and circuit photos; run hardware debug analysis; suggest net names; explain net connections
FirmwareGenerate Arduino sketches; compile firmware; upload to connected hardware; search Arduino libraries; list detected boards
For the complete list of every tool name, parameters, and description, see the AI Tool Actions Reference.

​
Action confirmation

When the AI selects a destructive tool — one that permanently deletes or clears data — the chat panel displays a confirmation card before the action executes. The card shows:
  • The tool name and a plain-English description of what will be deleted
  • An Approve button to proceed
  • A Cancel button to abort
The server enforces this protection independently of the UI. If a request arrives for a destructive tool without the confirmed: true flag, the server rejects it. This means the confirmation requirement cannot be bypassed by manipulating the client.

​
Action logging and undo

Every AI action that modifies design state is written to the ai_actions table in the database. You can review the full history in the Audit Trail view (switch_view → audit_trail). To reverse an AI action, either:
  • Type “undo that” or “undo the last action” in the chat — the AI will call the undo tool
  • Press Ctrl+Z directly in the workspace
  • Click the Undo button in the toolbar
Redo is available via the redo tool action or Ctrl+Y.

​
Explore further

AI Tool Actions Reference

Complete reference for all 82 tool actions, organized by module with parameters and descriptions.

Writing Effective Prompts

Prompting patterns, example prompts, and tips for getting consistent results from the AI.

Build docs developers (and LLMs) love

Get started for freeTalk to us