Skip to content

settings-tab.md

Latest commit

 

History

History
152 lines (109 loc) · 6.54 KB

settings-tab.md

File metadata and controls

152 lines (109 loc) · 6.54 KB
Raw

Settings Tab Reference

Screenshot

Purpose

The Settings tab manages LLM providers, MCP providers, SymGraph integration, the system prompt, and database paths. It is the central configuration hub for IDAssist.

LLM Providers

Provider Table

Lists all configured LLM providers:

Column Description
Name Unique provider name (e.g., "Claude", "GPT-5.3").
Model Model identifier (e.g., claude-sonnet-4-6, gpt-5.3-codex).
Type Provider type (see table below).
URL API endpoint URL.
Max Tokens Maximum tokens for requests.
TLS Whether TLS verification is enabled.

Provider Management Buttons

Button Description
Add Open a dialog to configure a new LLM provider.
Edit Modify the selected provider's settings.
Delete Remove the selected provider.
Test Send a test request to verify connectivity and credentials.

Active Provider

A dropdown selector to choose which LLM provider is used for all Explain, Query, and Actions operations. Only one provider is active at a time.

Reasoning Effort

Controls the extended thinking budget for supported models:

Level Token Budget Description
None Disabled No extended thinking. Standard response generation.
Low ~2K tokens Light reasoning for straightforward tasks.
Medium ~10K tokens Moderate reasoning for complex analysis.
High ~25K tokens Deep reasoning for the most complex tasks.

Extended thinking is supported by Anthropic models (Claude) and OpenAI reasoning models (o1).

Provider Types

Type Auth Method Streaming Tool Calling Notes
anthropic_platform API Key Yes Yes Direct Anthropic API
anthropic_oauth OAuth (browser) Yes Yes Browser-based auth flow
anthropic_claude_cli Local CLI Yes Yes Uses claude CLI binary
openai_platform API Key Yes Yes Direct OpenAI API
openai_oauth OAuth (browser) Yes Yes Browser-based auth flow
ollama None (local) Yes Model-dependent Self-hosted, no API key needed
litellm Proxy URL Yes Provider-dependent Multi-provider proxy

Provider Configuration Fields

When adding or editing a provider:

Field Description
Name Unique identifier for this provider.
Type Provider type from the table above.
Model Model ID (e.g., claude-sonnet-4-6, gpt-4o, qwen2.5-coder:32b).
URL API endpoint. Auto-filled for known provider types.
API Key Authentication credential. Encrypted in the settings database.
Max Tokens Maximum tokens per request.
Disable TLS Skip TLS certificate verification (for self-signed certs or local servers).

MCP Providers

MCP Provider Table

Column Description
Name Unique name for the MCP server.
Target HTTP endpoint for network transports, or the CLI command for stdio transports.
Enabled Whether this MCP server is active.
Transport Connection type (SSE, Streamable HTTP, or Stdio).

MCP Management Buttons

Button Description
Add Configure a new MCP server connection.
Edit Modify the selected MCP server's settings.
Delete Remove the selected MCP server.
Test Verify connectivity to the MCP server.

MCP Transport Modes

Use one of these transport styles when adding an MCP provider:

Transport Required Fields Notes
sse URL For MCP servers exposed over SSE, such as http://localhost:8000/sse.
streamablehttp URL For MCP servers exposed over streamable HTTP, such as http://localhost:8000/mcp.
stdio Command Starts a local MCP server process and talks to it over stdin/stdout.

For stdio providers, the dialog also supports:

  • Arguments: Command-line arguments. Enter either shell-style arguments or a JSON string array.
  • Working Directory: Optional process working directory.
  • Environment JSON: Optional JSON object of environment variables, for example {"API_KEY":"value"}.

On Windows, stdio MCP providers also require pywin32 in the same Python environment as IDA because the MCP SDK uses Windows Job Objects for child-process management.

SymGraph

The SymGraph section configures the remote service used by the SymGraph Tab.

Fields and actions:

  • API URL: SymGraph server URL for the hosted service or a self-hosted instance
  • API Key: Required for uploading binaries, importing shared analysis, and publishing revisions
  • Show / Hide: Toggle API-key visibility
  • Test: Verify connectivity and credentials

Refresh in the SymGraph tab can check whether a binary exists without an API key, but upload, import, and publish operations require one. IDAssist can only upload the raw binary when the original input file path is still available on disk.

System Prompt

A large text area for customizing the system prompt sent to the LLM with every request. The default prompt instructs the model to behave as a binary analysis assistant. Modify this to:

  • Focus the LLM on specific domains (malware, firmware, crypto)
  • Set output formatting preferences
  • Add standing instructions (always mention CVEs, always use C pseudocode, etc.)

Database Paths

Configure where IDAssist stores its local data:

Field Default Path Purpose
Analysis DB ~/.idapro/idassist/analysis.db Function summaries, security flags, graph data.
RLHF DB ~/.idapro/idassist/rlhf.db User feedback on LLM responses.
RAG Index ~/.idapro/idassist/rag_index/ Whoosh search index for uploaded documents.

Each field has a Browse button to select a custom path. The settings themselves are stored in ~/.idapro/idassist/settings.db.

Related Documentation

  • Getting Started — Initial provider setup walkthrough
  • Explain Tab — Uses the active LLM provider
  • Query Tab — Uses the active LLM provider and MCP servers
  • SymGraph Tab — Uses the SymGraph credentials configured here
  • RAG Tab — Uses the RAG index path configured here