In-App Agent

The Paracore Agent is a conversational AI built into the Desktop App. Describe what you want in plain language and the Agent generates C# using Paracore's fluent DSL, then executes it in Revit — with you in control at every step.

How It Works

The Agent uses Sovereign Code Generation. Traditional BIM AI agents need thousands of hardcoded tools (one per Revit API action). Paracore's Agent gets effectively one infinitely-capable tool: a C# compiler.

The Agent's Tools

Tool	Description	Requires Approval?
explore_revit_data	Silently inspects your model — discovers available categories, parameters, and element properties	No
search_schema	Fast parameter lookup from an in-memory cache — returns parameter names, storage types, and type/instance classification	No
read_extension_methods	Returns Paracore's fluent DSL reference so the Agent knows which methods are available	No
execute_dynamic_query	Generates the final C# script and pauses for your approval before running	Yes

The Agent can explore your model autonomously to understand its schema, but any code that mutates the model requires your explicit approval.

Thinking Steps

When the Agent works, you see exactly what it's doing before it produces code:

1. Explore — silently runs read-only C# to inspect categories, parameters, and element properties
2. Search — looks up parameter metadata from the in-memory schema cache
3. Read — consults the extension methods reference for the right fluent API
4. Generate — writes the C# script and presents it for your approval

Each step appears in the chat as it happens, giving you full visibility into the Agent's reasoning process.

Setup

1. Go to TopBar → Settings → LLM Settings
2. Select a provider: DeepSeek, Google Gemini, OpenAI, Anthropic, or OpenRouter
3. Choose a model and enter your API key
4. Click Save

No sign-in is required — click Continue Offline and the Agent is fully usable.

Chat Sessions

The Agent supports multiple named chat sessions, each with independent history persisted across app restarts:

• New Session: Start fresh with a clean conversation — useful for switching between different tasks or models.
• Switch Sessions: Jump between conversations without losing context.
• Delete Sessions: Remove old conversations when they're no longer needed.
• Token Tracking: Each session shows cumulative token usage so you can monitor API costs.

Human-in-the-Loop

When the Agent proposes code that would modify the Revit model, a HITL modal appears showing the full C# code:

• Approve: The code runs immediately in Revit.
• Reject: The code is discarded. You can refine your request and try again.
• Copy: Copy the code to use elsewhere (REPL, gallery script, etc.).
• Self-Correction: If execution fails, the Agent reads the error and rewrites the code automatically (up to 3 retries).

Stale approvals (from earlier turns) are marked SKIPPED to prevent accidental out-of-order execution.

Mid-Stream Abort

If you send a new message while the Agent is generating, the current generation is aborted cleanly and your new message is preserved. No lost context, no orphaned tool calls.

Multi-LLM Support

The Agent works with any supported provider. Switch models mid-session without losing conversation history. Each provider/model combination is independent — you can use DeepSeek for quick queries and switch to a more capable model for complex tasks.

---

Next: Connect external AI tools with the MCP Server.