Selvedge
Your coding agent's memory for decisions and dead ends
FreeCLIOpen SourceAI2,179 impressions#8 of its week5 comments
Comments
>log in to commentMaker here — what's one approach your coding agent keeps suggesting after you've already rejected it? This 45-second demo shows a rejection saved in one session and retrieved in a fresh session from the same local database: https://youtu.be/madtsGupCcM The lookup is explicitly prompted. Selvedge recalls what you or your agent saved; it doesn't automatically capture every decision. Try the built-in terminal demo (Python 3.10+ and uv): uv tool install --upgrade selvedge selvedge demo That demo saves a rejection and retrieves it through a fresh connection to a temporary database. Your project stays untouched. Tell me which agent you use and one non-sensitive rejected approach you'd want it to look up. Failed lookups and confusing setup are useful feedback too. If you find Selvedge useful, an upvote is welcome. — Mason
Thanks, Vinny — glad the local-first approach resonates! If you try a fresh-session lookup, I'd love to hear which agent you used and how it went.
- Vinny Vidovich· 9d ago
Something that works. Creator on to something with this. Fact it keeps notes in a local file 🤯
Thanks, Matin — appreciate the support!
- matinGhaffari· 10d ago
Honestly such a game changer
New session. Same rejected idea? Your agent can read the code. It may not know why you rejected an approach last week.
Keep the why with your project. Selvedge stores decisions, rejected approaches, and when to reconsider them. Save a real decision in one session; retrieve its reason in the next through MCP or the CLI.
Local SQLite. MIT open source. No account, subscription, or LLM calls by Selvedge.
Bring one repeat bug
Your agent keeps revisiting a bug or an approach you already rejected? Bring one example to a small, free pilot. I'll help you record the reason and check whether a fresh coding session can retrieve it. Failed lookups and confusing setup are useful feedback too.
Talk to Mason and join the pilot →
Reply with your coding agent and a short, non-identifying project label. Participation is voluntary; no testimonial is required. The discussion is public, so keep private code and customer information out and read the notes-sharing details first.
See it work
Watch the 45-second agent demo → One coding agent saves a rejected approach. A fresh agent retrieves its reason through MCP. Two separate sessions, one local database. Actual captured output; timing edited and retrieval explicitly prompted.
Inspect the tool calls and method →
Founder feedback
“Selvedge has been great for our team at Vora.”
— Matin Ghaffari, co-founder of Vora Health. Read his Product Hunt review →
Try it in your terminal
uv tool install --upgrade selvedge
selvedge demoThe demo saves a rejection and retrieves it through a fresh connection to a temporary database. Your project stays untouched. Requires Python 3.10+ and uv; pip instructions are in the quickstart.
Set up your agent → · Explore the source →
Selvedge remembers what you or your agent explicitly save. Installed instructions guide the agent; they do not guarantee every decision is captured.
Selvedge keeps project decisions, rejected approaches, and their reasons available to your next AI coding session through MCP and the CLI.
- for
- Developers using AI coding assistants who need to track and reuse decision context.
- pricing
- free
- license
- MIT
- v0.3.15Selvedge v0.3.1522 days ago
- v0.3.14Selvedge v0.3.14a month ago
- v0.3.13Selvedge v0.3.13a month ago
Key features
- Local SQLite storage — Decisions are saved in a project-local .selvedge/selvedge.db file without external services.
- Agent-agnostic integration — Connects to Claude Code, Codex, Copilot, Cursor, Gemini CLI and Windsurf through MCP. Setup and lifecycle-hook support vary by client.
- Explicit decision recording — Save what was decided, which approaches were rejected, why, and the conditions for reconsidering them.
- CLI and MCP access — Retrieve stored decisions through a Python CLI or MCP calls in new sessions.
- Zero-account, open source — MIT-licensed, no subscription, and makes no LLM calls itself.
- Project-portable database — Carry the SQLite file to another machine or export a readable decision log.
Use cases
- Recall why a specific API key storage method was rejected before refactoring.
- Share decision history between multiple AI agents working on the same codebase.
- Mark when a decision should be reconsidered, then query decisions due for review.
Selvedge FAQ
Do I need an account or subscription to use Selvedge?+
No, Selvedge is MIT-licensed, requires no account, and has no subscription fees.
Does Selvedge make any external LLM calls?+
Selvedge itself makes no LLM calls. Your connected coding agent follows its own provider's data policy.
What Python version is required?+
Python 3.10 or newer. Install with uv, or use pip in a virtual environment.
How does Selvedge integrate with my AI coding agent?+
Run selvedge setup for your agent, review the configuration changes, and restart the client. Your agent can then use MCP tools to explicitly save decisions and retrieve them in later sessions.
Can I move the decision database to another machine?+
Yes, the SQLite file is project-local, so you can copy the .selvedge folder to another environment.
Is there any telemetry or data sent back to Selvedge?+
Product usage telemetry is off by default. The website records anonymous aggregate interactions and offers optional install confirmations. See https://selvedge.sh/privacy/ and the linked product documentation for details.
Edited by the makers, based on selvedge.sh · Sep 29, 2026. Details may change; check the official site.
