Manages token budget for LLM context window. Divides the context window into sections with reserved allocations: - System prompt: Fi

Orchestrates context management for LLM agent loops. Manages: - Token budget tracking across prompt sections - Tool output buffe

Manages conversation blocks with summarization support. Maintains a buffer of conversation blocks and provides methods to: - Add new mes

Reusable dialog for selecting functions for analysis.

OpenAI-compatible LLM client for memory retrieval tasks.

Represents a single turn in the conversation. Attributes: role: Message role (user, assistant, tool, system) content: Me

Master index of all classified functions for a binary.

LLM-guided retrieval over the memory node hierarchy. The client is intentionally duck-typed so callers can provide any object with the memory

Token counts for each section of the assembled prompt. Used for tracking cache-stable vs cache-volatile portions.

Tracks progress across LLM classification batches.

Summary of a batch of conversation blocks. Attributes: summary_by_action: Legacy grouped summary data when returned by the model

Summarized block of conversation turns. Replaces multiple turns with a single summary message. Attributes: turn_range:

Registry for tags created by the LLM that aren't in the configured taxonomy.

A single indexed function.

Manages tool call history with intelligent truncation. Maintains a ring buffer of tool call records and applies progressive truncati

Engine for local embeddings and vector quantization using Qwen and TurboQuant.

Form wrapper that includes controls and the AI decompilation viewer.

Hooks into Hex-Rays events to detect when the user changes functions.

Hooks into Hex-Rays events to detect when the user changes functions.

Hooks into Hex-Rays events to detect cursor position changes for scrolling.

Custom viewer for AI decompilation display.

Custom viewer for AI decompilation display.

Form wrapper that includes controls and the AI deobfuscation viewer.

Hooks into Hex-Rays events to detect cursor position changes for scrolling.

Custom viewer for AI deobfuscation display.

Result of ContextManager.assemble_prompt(). Contains both the messages for the LLM API and the updated conversation history with summari

Manages multiple asyncio worker threads for concurrent async operations.

A thread for running the asyncio event loop.

C/C++ syntax highlighter that closely follows the hilite.html implementation.

C/C++ syntax highlighter that closely follows the hilite.html implementation.