Memory banks are isolated containers that store all memory-related data for a specific context or use case. Each bank holds memories, documents, entities, relationships, and directives. Banks are completely isolated — memories in one bank are never visible to another. You do not need to pre-create a bank; Hindsight creates it automatically with default settings when you first use it.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/vectorize-io/hindsight/llms.txt
Use this file to discover all available pages before exploring further.
Create or update a bank
Request parameters
Your tenant identifier. Use
default for single-tenant deployments.Unique identifier for the bank. Alphanumeric and hyphens.
A first-person narrative that provides identity and framing context for
reflect. The agent uses this to ground its reasoning and apply a consistent perspective. Example: "You are a senior engineering assistant. Always ground answers in documented decisions. Be direct and precise."How skeptical vs. trusting the bank is when evaluating claims during
reflect. Scale 1–5. 1 = trusting (accepts information at face value), 3 = balanced (default), 5 = skeptical (questions and doubts claims).How literally to interpret information during
reflect. Scale 1–5. 1 = flexible (reads between the lines), 3 = balanced (default), 5 = literal (takes things exactly as stated).How much to weight emotional context when reasoning during
reflect. Scale 1–5. 1 = detached (focuses on facts and logic), 3 = balanced (default), 5 = empathetic (considers emotional context).Response fields
The bank identifier.
ISO 8601 timestamp of when the bank was created.
ISO 8601 timestamp of the most recent update.
Example
Get a bank
Response fields
The bank identifier.
The bank’s reflect mission, if set.
Current skepticism setting (1–5).
Current literalism setting (1–5).
Current empathy setting (1–5).
ISO 8601 creation timestamp.
ISO 8601 last-updated timestamp.
Example
List banks
Example
Delete a bank
Example
Bank configuration
Bank configuration fields (extraction mode, missions, observations, disposition, recall budgets, entity labels) are managed via a separate config API endpoint, not thePUT /banks/{bank_id} call. This lets you change operational settings independently from the bank’s identity.
Update configuration
A plain-language description of what this bank should pay attention to during extraction. Injected into the extraction prompt alongside built-in rules. Example:
"Always include technical decisions and API design choices. Ignore meeting logistics and greetings."Controls how aggressively facts are extracted.
"concise" — selective, only facts worth remembering long-term. "verbose" — captures more detail per fact, slower and uses more tokens. "custom" — use your own extraction rules via retain_custom_instructions.Custom extraction prompt. Only active when
retain_extraction_mode is "custom". Replaces the built-in extraction rules entirely.Maximum number of characters per chunk when splitting content for fact extraction.
Toggles automatic observation consolidation on or off.
Defines what this bank should synthesize into durable observations. Replaces the built-in consolidation rules. Example:
"Observations are stable facts about people and projects. Always include preferences, skills, and recurring patterns."Selects how the recall
budget parameter maps to the internal retrieval limit. "fixed" — uses recall_budget_fixed_<level> independent of max_tokens. "adaptive" — scales retrieval breadth with the requested output size.Defines a controlled vocabulary of
key:value classification labels extracted at retain time. Each entry is a label group with key, description, type ("value", "multi-values", "text", or "map"), values (allowed enum values), optional, and tag (when true, extracted labels are also written as tags).Allowlist of MCP tool names enabled for this bank. Set to
null to allow all tools.Example
Get configuration
Reset configuration
Bank profile
Response fields
The bank identifier.
Total number of memory facts stored in the bank.
Total number of documents.
Total number of entities in the knowledge graph.
Total number of consolidated observations.
Example
Directives
Directives are hard rules that the agent must follow during reflect operations. Unlike disposition traits which influence how the agent reasons, directives are explicit instructions that are always enforced.Directives only affect the
reflect operation. They are injected into prompts and the agent is required to comply in all responses.Create a directive
Human-readable name for the directive.
The directive content or rules. Example:
"Never share personal data with third parties." or "Always respond in formal English."Priority level. Higher values are injected first into the prompt.
Whether the directive is currently active.
Tags for filtering which directives apply to a given reflect call.
