Bank Configuration

Each agent gets a bank config file — a JSON5 template that defines how Hindsight processes that agent's memories.

Bank Config File Reference

Field	Type	Scope	Description
retain_mission	string	Server	Guides fact extraction during retain
observations_mission	string	Server	Controls observation consolidation
reflect_mission	string	Server	Prompt for reflect operations
retain_extraction_mode	string	Server	Extraction strategy (concise, verbose)
disposition_skepticism	1-5	Server	How skeptical during extraction
disposition_literalism	1-5	Server	How literally statements are interpreted
disposition_empathy	1-5	Server	Weight given to emotional content
entity_labels	EntityLabel[]	Server	Custom entity types for classification
directives	{name,content}[]	Server	Standing instructions for the bank
retain_strategies	Record	Server	Named extraction strategies
retain_default_strategy	string	Server	Fallback strategy when no named strategy is passed
retain_chunk_size	number	Server	Text chunk size for processing
retain	RetainRouting	Routing	Topic-based strategy routing (plugin-side)
retainTags	string[]	Tags	Tags added to all retained facts
retainContext	string	Tags	Source label for retained facts
recallFrom	string[]	Multi-bank	Banks to query (parallel recall)
sessionStartModels	config[]	Session	Mental models loaded at session start
reflectOnRecall	boolean	Reflect	Use reflect instead of recall
reflectBudget	low|mid|high	Reflect	Reflect effort level

Server-side fields are synced to Hindsight via hindclaw apply. Routing and behavioral fields are used by the plugin at runtime.

Example Bank Config

// .openclaw/banks/agent-1.json5
{
  // Server-side — synced to Hindsight
  "retain_mission": "Extract strategic decisions, cross-departmental patterns.",
  "reflect_mission": "You are the strategic advisor. Challenge assumptions.",
  "disposition_skepticism": 4,
  "disposition_literalism": 2,
  "disposition_empathy": 3,
  "entity_labels": { "$include": "./agent-1/entity-labels.json5" },
  "directives": [
    { "name": "cross_dept_honesty", "content": "Flag contradictions between departments explicitly." }
  ],

  // Named strategies — different extraction rules per context
  "retain_strategies": {
    "deep-analysis": { "$include": "./agent-1/deep-analysis.json5" },
    "lightweight":   { "$include": "./agent-1/lightweight.json5" }
  },

  // Strategy routing — which topics use which strategies
  "retain": {
    "strategies": {
      "deep-analysis": { "topics": ["280304"] },
      "lightweight":   { "topics": ["280418"] }
    }
  },

  // Multi-bank recall
  "recallFrom": ["agent-1", "agent-2", "agent-3"],
  "recallBudget": "high",
  "recallMaxTokens": 2048
}

Named Retain Strategies

Route memory behavior per conversation context. Each Telegram topic can use a different retain strategy with its own extraction rules.

{
  "retain_strategies": {
    "deep-analysis": {
      "retain_extraction_mode": "verbose",
      "retain_mission": "Extract every decision, risk, and opportunity in full detail."
    },
    "lightweight": {
      "retain_extraction_mode": "concise",
      "retain_mission": "Only keep hard facts — dates, numbers, action items."
    }
  },
  "retain": {
    "strategies": {
      "deep-analysis": { "topics": ["12345"] },
      "lightweight":   { "topics": ["67890"] }
    }
  }
}

Example Scenarios

Strategic advisor — one agent, three conversation contexts:

Topic	Strategy	What happens
"Strategy"	deep-analysis	Every decision, risk, and opportunity extracted with verbose detail
"Daily updates"	lightweight	Only hard facts — dates, numbers, action items
"Weekly review"	(no strategy)	Recall only, review conversation not retained

Health agent with data boundaries:

Topic	Strategy	What happens
"Fitness log"	training	Extracts sets, reps, PRs, recovery scores
"Medical"	(disabled)	No memory interaction — strict boundary
"Sleep"	wellness	Tracks sleep patterns, recovery scores

$include Directives

Split large configs into manageable files. Resolved recursively, relative to the containing file:

// Main bank config
{
  "entity_labels": { "$include": "./agent-1/labels.json5" },
  "retain_strategies": {
    "detailed": { "$include": "./agent-1/detailed-strategy.json5" }
  }
}

.openclaw/banks/
├── agent-1.json5                  <- main bank config
├── agent-1/
│   ├── labels.json5               <- entity label definitions
│   ├── detailed-strategy.json5    <- strategy: verbose + custom labels
│   └── quick-strategy.json5       <- strategy: concise extraction

Limits: max depth 10, circular reference detection, paths relative to containing file.