Hindsight API

MCP Active

Hindsight: Agent Memory That Learns

Purpose

To provide AI agents with a sophisticated, human-like memory system that enhances their ability to store, recall, and reason about information over time.

Features

Temporal and semantic memory retrieval
Entity graph for relationship tracking
Configurable disposition traits for opinion formation
Support for multiple LLM providers
REST API, CLI, and MCP server interfaces

Use Cases

Building persistent memory for long-running AI agents
Enabling agents to reason about past events and user preferences
Integrating memory capabilities into existing AI workflows via MCP
Developing AI assistants that learn and adapt over time

Non-Goals

Acting as a general-purpose database
Providing direct LLM inference capabilities
Replacing the core agent execution logic

Workflow

Initialize memory engine
Create or select a memory bank
Store (retain) facts, events, or opinions
Recall memories based on queries
Reflect on memories to form new opinions or reason

Practices

Data management
AI memory architecture
API design
Security best practices

Prerequisites

PostgreSQL (or embedded pg0)
LLM API key and provider configuration
Python 3.11+

Maintenance

warning:Dependency ManagementThe pyproject.toml lists numerous pinned dependencies and transitive dependency fixes, but lacks explicit measures like Dependabot configuration for automated updates and vulnerability checks.

Code Execution

info:LoggingThe application uses standard Python logging, with configurable levels, but doesn't appear to implement a dedicated local audit file for destructive actions.

Execution

info:Pinned dependenciesDependencies are pinned in `pyproject.toml`, but lockfiles for transitive dependencies are not explicitly managed or checked for vulnerabilities.