Sugar
Persistent memory for AI coding agents
The Challenge
AI coding agents like Claude Code and Aider are powerful, but they forget everything between sessions. Every new conversation starts from scratch - agents re-learn your project conventions, repeat past mistakes, and have no memory of decisions made yesterday. Development teams waste time re-explaining the same context over and over.
The Solution
Sugar gives AI coding agents persistent memory that spans sessions. Agents store and recall project conventions, architectural decisions, error patterns, and codebase context through a built-in MCP server. The next time an agent picks up a task, it already knows your project like a team member who has been there for months.
AI coding agents can write code, but they forget everything between sessions. Every new conversation starts from scratch - agents re-learn your conventions, repeat mistakes they already solved, and have no memory of decisions made yesterday. Teams waste time re-explaining the same context over and over.
Sugar fixes that with typed, persistent memory that spans sessions. Not a blob of text - seven distinct memory types (decision, preference, file context, error pattern, research, outcome, guideline) each with different retrieval behavior and TTL logic. An architectural decision is stored and recalled differently than a coding preference or an error pattern. Your agents remember your project like a team member who has been there for months.
Vector search uses sentence-transformers locally - a 90MB model that runs on your machine, no external API calls. When you or the agent stores a memory, it gets embedded and indexed for semantic retrieval. Search by meaning, not keywords. Falls back to FTS5 keyword search if you skip the embeddings install.
We built a custom MCP server into Sugar so that AI agents interact with memory directly through the Model Context Protocol. Agents store learnings, search past decisions, and retrieve project context at the start of every session - all through a standardized interface. This is the same type of MCP server we build for clients who want their AI systems to connect with internal tools and data sources.
Sugar is open source, works with Claude Code, OpenCode, Aider, and Goose, and is used by development teams who want their AI agents to get smarter over time instead of starting from zero every session.
Key Features
How AI Powers Sugar
Agents That Remember
Persistent semantic memory means agents do not start from scratch each time. They recall project conventions, past decisions, and patterns from previous sessions - like a team member who has been on the project for months.
Custom MCP Server Built In
We built a production MCP server into Sugar so AI agents store and retrieve memory through the Model Context Protocol. Agents recall context, store new learnings, and search past decisions - all through a standardized interface. This is the same kind of MCP server we build for clients.
Scoped Memory
Project-level memory keeps context separate across codebases. Global memory shares cross-project knowledge like coding standards and infrastructure patterns. Agents always get the right context for the task at hand.
Vector Search with Embeddings
Agents search memory by meaning, not just keywords. Sugar generates embeddings with sentence-transformers and runs vector similarity search locally - no external APIs. Ask about "how we handle auth" and get back the architectural decision from three weeks ago, even if the words do not match exactly.
Tech Stack
Need something like this built?
RoboticForce builds production AI systems, MCP servers, and APIs. We ship working software, not slide decks.
Start a project