Vibeleaderboard
Index / tool
Visit github.com
Category
AI Agents
Rank
No. 2115Tools index

Previous survey · No. 2031 ·

Type
TOOL
Use case
Coding
Interfaces
Agent Skill / Plugin
GitHub
4 stars
Date

About

Singularity gives coding agents memory of past work. It learns from finished sessions where each kind of change goes in a repo, how it is checked, and which mistakes occurred, then a hook hands matching workflows to the agent at the start of the next task. On excalidraw, a four-change task dropped from 1.9M to 423k tokens, though small tasks and one-off bug fixes showed no meaningful gain.

At a glance

Singularity records what a coding agent did around a commit and retrieves a relevant past change for a later task. It is worth trying on repositories where agents repeat maintenance work and the old checks are hard to rediscover from Git alone.

Good fit

Developers whose coding agents revisit similar tasks in the same repository.

Consider something else when

Skip it if most tasks are one-offs or your existing project instructions already point agents to the right checks.

Start here

In a scratch repository, have an agent make a small commit after running a test. Inspect the recorded change, then start a related task and see whether the earlier commit and test are surfaced.

Cost and requirements

MIT-licensed. The project documents macOS, Linux, WSL and Git Bash setup. Its benefit depends on compatible agent sessions and changes it can associate with commits.

Limits to know

The project says stored memory stays on the machine. Records include commands and task context, so inspect what was stored before reusing it. The reported results used Claude Code with its own automatic memory disabled.

Related work

Intel on Singularity

More in Intel

Tags

coding-agentmemoryclaude-codetokenshookscodex

Tech Stack

Node.js

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.