Rorschach

Logging observability · log output and level settings

Log file path JSONL output file, relative to project root

Minimum log level Events below this level are suppressed

Flush interval (ms) How often buffered log entries are written to disk

Metrics observability · actor telemetry collection

Metrics interval (ms) How often actor metrics snapshots are emitted

Enable metrics collection

Collect per-actor processing time and message counts

Chat cognitive · language model and reasoning

Model

System prompt Persona and instructions given to the model before each conversation

History window Number of messages kept in context per session

Enable reasoning

Stream thinking tokens for supported models

Reasoning effort Token budget allocated for thinking

Vision cognitive · image analysis model

Vision model Model used by the analyze_image tool

Audio tools · speech-to-text and text-to-speech

Audio model Model used for transcription and speech synthesis

TTS voice Voice persona for text-to-speech output

Tools tools · bash execution and web search

Bash working directory Default working directory for bash and file tools

Web search result count Maximum number of results returned per web search

Memory memory · knowledge graph and consolidation

Knowledge graph path Directory where the graph database is stored

Memory model Model used for memory consolidation and recall

Embedding model Model used to generate vector embeddings for the knowledge graph

Embedding dimensions Vector size produced by the embedding model

Consolidation interval (ms) How often background memory consolidation runs

Reflection interval (ms) How often the memory reflection pass runs (default 7 days)

Notebook notebook · notes, journal, todos, and tracker

Notebook directory Root directory where notes, journal entries, and attachments are stored

Agent model Model used by the notebook sub-agent to handle tool requests

Consolidation interval (ms) How often the notebook consolidation pass runs (default 7 days)

Max tool loops Maximum tool call iterations per notebook agent request