Skip to main content

Introduction to Hindsight Cloud

Hindsight Cloud is the managed cloud service for Hindsight, a state-of-the-art memory system designed specifically for AI agents.

The Problem​

AI agents forget everything between sessions. Without persistent memory, agents can't maintain context across conversations, learn from past interactions, or build lasting relationships with users. Every conversation starts from zero.

Hindsight solves this by providing agents with a sophisticated memory system that goes far beyond simple storage and retrieval.

What is Hindsight?​

Hindsight enables AI agents to remember, learn, and reason through three core operations:

  • Retain - Store information in dedicated memory banks, automatically extracting facts, entities, and temporal data
  • Recall - Search and retrieve memories using multiple parallel strategies
  • Reflect - Perform agentic reasoning over retrieved memories, guided by a bank's mission, directives, and disposition traits

Key Concepts​

Memory Banks​

A Memory Bank is a dedicated memory space for a specific agent or context. Each bank maintains its own:

  • Stored memories across different types
  • Entity relationships and graph connections
  • Mission, directives, and disposition traits that guide reasoning
  • Search indices for fast retrieval

Memory Hierarchy​

Hindsight organizes knowledge into a layered hierarchy, from raw facts up to curated summaries:

LevelDescriptionExample
World FactsObjective facts received from external sources"Alice works at Google"
Experience FactsThe agent's own actions and interactions"I recommended Python to Bob"
ObservationsAutomatically synthesized knowledge — new facts are analyzed and consolidated into observations with evidence tracking"User is growing comfortable with async Python"
Mental ModelsPre-computed, curated summaries for common queries — the fastest path to consistent answers"Team communication best practices"

During reasoning, Hindsight checks sources in priority order: Mental Models → Observations → Raw Facts. After each retain call, observation consolidation runs automatically in the background, analyzing new facts against existing observations and creating or refining them as patterns emerge.

Mission & Directives​

Each memory bank can be configured with a mission and directives that shape how the agent reasons during reflect operations:

  • Mission - A natural language identity statement that tells Hindsight what knowledge to prioritize and provides context for reasoning. For example: "I am a research assistant specializing in ML. I prefer simplicity over cutting-edge."
  • Directives - Hard rules the agent must follow during reflect — guardrails and compliance requirements that must never be violated. For example: "Never recommend specific stocks" or "Always cite sources."

The mission provides the interpretive lens, directives enforce boundaries, and disposition traits modulate reasoning style. Together they define how a bank synthesizes and communicates stored knowledge during reflect.

TEMPR Retrieval​

Unlike systems that rely solely on semantic search, Hindsight uses the TEMPR retrieval strategy—four parallel search methods that work together:

MethodWhat it finds
SemanticConceptually similar memories
Keyword (BM25)Exact term matches
GraphEntity-connected memories
TemporalTime-based reasoning ("last week", "in March")

This multi-strategy approach handles queries that semantic search alone cannot answer, like "What did Alice tell me last spring?"

Disposition Traits​

Memory banks can have disposition traits that influence how the agent reasons during reflect operations:

  • Skepticism - How readily the agent accepts new information (trusting ↔ skeptical)
  • Literalism - How flexibly the agent interprets statements (flexible ↔ literal)
  • Empathy - How much weight the agent gives to emotional context (detached ↔ empathetic)

Hindsight Cloud Features​

Hindsight Cloud adds enterprise-ready capabilities on top of the open-source Hindsight core:

  • Managed Infrastructure - No servers to deploy or maintain
  • API Integration - RESTful API with Python and TypeScript SDKs
  • Team Management - Invite team members with role-based access control
  • Usage Analytics - Real-time dashboards showing consumption
  • Credit System - Usage-based pricing with credits for API operations

Token-Based Pricing​

Usage is measured in tokens:

  • Retain tokens - Consumed when storing new memories
  • Recall tokens - Consumed when searching/retrieving memories
  • Reflect tokens - Consumed when generating AI-powered insights
  • Mental Model tokens - Consumed when creating, refreshing, or retrieving mental models

Getting Started​

Ready to give your AI agents persistent memory? Head to the Getting Started guide to create your first memory bank.