The shortlist
7 projects
The wider reading list: projects that stood out across the week, in their own words. Positions reflect the shortlist, not GitHub rank.
Also worth watching
Fast risers from off the boards — worth a look anyway.
The wider reading list: projects that stood out across the week, in their own words. Positions reflect the shortlist, not GitHub rank.
Fast risers from off the boards — worth a look anyway.
The weekly Star
A project by vectorize-io
An agent memory system that lets LLM agents learn over time by retaining, recalling, and reflecting on experiences, not just chat history.
You know the ritual. New session, and your agent has never met you. So you type the setup again: the project, the preferences, the decision from last Tuesday that keeps not carrying over. The agent nods along, because that is all it can do. It has amnesia in both directions, and you are the one paying for it in typing.
Hindsight is memory for your agent. Hook it up once, and it listens from then on. It keeps the stuff worth keeping, who you are, what you're building, what changed, and next time you open a session, the relevant parts are already in the room before your first word lands. To the agent, you're someone it has known for months.
Stars at each capture
Rank on each board
Two habits matter, and they're the whole difference from the neighbors.
It doesn't store your sessions. It stores what a session established, plus the sentence you said it in. When your deployment cutoff moves to Fridays, that becomes a memory with a receipt attached. Six months later, when the agent confidently tells someone the cutoff is Wednesday, you can go look at what the memory is built on instead of arguing with it.
And memories age instead of getting overwritten. Prefer a tool in March, dump it in August, and both chapters survive, the newer one in charge. Ask what you use and you get the current answer. Ask what you used to use, and it still knows. Most systems just replace the old entry and lose the trail, which is fine until you need the trail.
For coding agents this is a ten-minute job. Claude Code, Cursor, Codex, Cline, Copilot, OpenCode, Zed, and most of the others have an integration sitting in the tree. Building your own thing instead? The entire system is a Python package, no server, nothing to feed:
pip install hindsight-allThe brain it borrows is a model you already have. Your existing Claude, Cursor, or Copilot subscription works, no new API key. Fully local through Ollama or LM Studio if your conversations should stay on your machine. There's even an Obsidian bridge, for when your notes are where your life already lives.
Agent memory is a crowded shelf, and the boxes sound identical until you read the side panels.
Mem0 is the default answer, and the default is not wrong. One command, one shared memory pool across Claude and Cursor and the rest. If you just want your agents to remember you with minimum ceremony, start there or here and you'll be fine either way.
Zep's open engine, Graphiti, keeps facts in a graph you host yourself, database and all. Strongest in the field at questions with time in them. It's the right pick when memory is infrastructure and somebody's job includes tending it.
Letta isn't something you attach. It's a whole agent framework where memory is one organ in a larger body, so adopting it means building your agent on Letta from day one. Powerful. Much bigger decision.
Cognee is happiest turning documents into a knowledge base: your wiki, your code, your archive. Less interested in you, more interested in your library.
Supermemory plugs in easily, runs fast, and doesn't spend much thought on what it stores.
So Hindsight's case comes down to those two habits from earlier: receipts, and aging instead of vanishing. If neither does anything for you, the shelf has easier options.
Two symptoms mean yes. You repeat yourself: same setup, same preferences, every Monday, to a new session. Or you're fragmented: your terminal agent, your phone assistant, and your editor each know a different piece of you, and none of them the whole.
If your agent work is occasional, one-off tasks, skip the whole category. A short instructions file does the job and costs nothing. The day the repetition starts to hurt, land on Hindsight or Mem0, and the choice between them is simple: do you want your memories to show their work?
Open-source SDK for building spreadsheet, document, and presentation experiences embedded in web apps or headless Node.js services.
Days on board: 5 daily · 1 weekly · 0 monthly
An agent memory system that lets LLM agents learn over time by retaining, recalling, and reflecting on experiences, not just chat history.
Days on board: 4 daily · 4 weekly · 0 monthly
A declarative orchestrator for autonomous agent workloads in sandboxed clusters, driven by Kubernetes-style YAML and a kubectl-like CLI.
Days on board: 3 daily · 0 weekly · 0 monthly
Claude plugins and agent templates bundling finance skills, slash commands, and data connectors for banking, research, and fund ops.
Days on board: 3 daily · 5 weekly · 3 monthly
A hosted tool registry and proxy that lets agents call 3,000+ external API endpoints, plus shared team tools, with one base URL and token.
Days on board: 2 daily · 4 weekly · 4 monthly · specialist pick
LLM-powered knowledge framework for enterprise document understanding, RAG Q&A, agent reasoning, and auto-built wiki knowledge bases.
Days on board: 0 daily · 6 weekly · 6 monthly · specialist pick
Terminal-based agentic coding tool that understands your codebase and handles tasks, code explanations, and git via natural language.
Days on board: 0 daily · 6 weekly · 0 monthly · specialist pick
Ship the product your agent just built — hosting, database, domain, email, and payments on your own accounts, planned and applied by an open skill with a zero-dependency CLI — no vendor account, no backend, no telemetry.