stm.example.jsonc

{
  // Session memory plugin on/off
  "enabled": true,

  // Model used by the summarizer
  "memoryModel": "opencode/minimax-m2.5-free",

  // clean | active
  "summarizerMode": "clean",

  // If true, clean-mode failures will call active-session prompt fallback.
  // Keep false to prevent side-session prompts from appearing in desktop chat.
  "cleanFallbackToActiveSession": false,

  // If true, include AGENTS.md content in the first memory update prompt per session
  "includeAgentsMdOnFirstUpdate": false,

  // If true, inject parent session memory into sub-agent (fork) sessions
  "injectInSubagents": true,

  // Number of retries for clean side-session summarizer failures (0 = no retry)
  "sideSessionRetries": 1,

  // Interval at which to inject memory summarization into chat (every N prompts from user)
  "remindEveryN": 4,

  // Max chars stored in session memory markdown
  "maxMemoryLength": 10000,

  // Max chars from recent conversation sent to summarizer
  "maxUpdateInputLength": 20000,

  // Max number of recent messages processed per update cycle (limits look-back on stale checkpoints)
  "maxDeltaMessages": 200,

  // Collapse consecutive assistant messages between user turns into the last reply
  // true = lower token usage, may drop intermediate reasoning
  // false = keep every assistant turn (default)
  "collapseAssistantBursts": false,

  // Debounce for scheduled updates (ms)
  "debounceMs": 1200,

  // Debug logging
  "debug": false,

  // Max JSONL lines kept in .opencode/memory/session-memory.log
  "logMaxLines": 300,

  // Directory where memory files, checkpoints, and logs are stored (relative to project root)
  "memoryDir": ".opencode/memory",
}