Jaixii | Building Engraphis

Building @Engraphis, local-first memory for coding agents. I write about agent context, open source, and developer tools.

[Jaixii | Building Engraphis @CodingDevTools]

@AlanBin44785396 Exactly. Rejected options are the expensive part: they carry why a path was ruled out, and summaries drop that first. A useful handoff is a decision log with constraints and a tried/rejected list, not a status paragraph. Codex notes: https://t.co/VTV8pgYtDE

[Jaixii | Building Engraphis @CodingDevTools]

A handoff can pass the summary and still lose the state. The useful handoff is decisions, constraints, evidence, and open questions - not a re-summary of the tree. Checklist: https://t.co/CEd7RpSW1e (I maintain Engraphis.)

[Jaixii | Building Engraphis @CodingDevTools]

Cursor rules do not remember why the code changed. We published a Cursor memory guide: scoped decisions, temporal history, and a 12-point checklist.
https://t.co/PTOLYDTReY

[Jaixii | Building Engraphis @CodingDevTools]

I turned this into a 12-question checklist for evaluating memory tools:

https://t.co/c6qrspUo7P

Use it on any tool, including mine.

[Jaixii | Building Engraphis @CodingDevTools]

Your coding agent can be right about the code and wrong about the decision.

A demo memory says JWT. A newer one says PASETO replaced it. Without supersession, a new session retrieves both and confidently suggests the old one.

That is a memory problem, not a model problem.

[Jaixii | Building Engraphis @CodingDevTools]

A better pattern:

1. Write decisions as small claims, not transcripts.
2. Tag scope, source, and timestamp.
3. Supersede old claims instead of appending.
4. Retrieve a small relevant set.
5. Keep a short working summary.

That is the difference between storage and memory.

[Jaixii | Building Engraphis @CodingDevTools]

I build Engraphis, a local-first memory engine for coding agents. The checklist is the useful part even if you choose something else.

What has your agent forgotten that you wish it remembered?

[Jaixii | Building Engraphis @CodingDevTools]

Your coding agent forgets. Engraphis gives Claude Code, Cursor, and Codex one local memory store: scoped, temporal, cited or abstain. Apache-2.0. Three commands to a cited recall:
https://t.co/eSjC3cCdFK

[Jaixii | Building Engraphis @CodingDevTools]

5% monthly churn sounds manageable. It’s actually 46% annual churn.

The bad news: nearly half your customer base still vanishes every year.

Calculate your compounding churn & revenue leakage in 15 seconds (free, no signup):

https://t.co/kTYuz4dIC8

#SaaS #SaaSMetrics

[Jaixii | Building Engraphis @CodingDevTools]

@danikkk_wqs Try https://t.co/VlvjyLbRB7 for an all in one optimal agentic memory system

[Jaixii | Building Engraphis @CodingDevTools]

Rejected approaches are evidence. Workarounds are temporary. The failure is when the workaround outlives the bug and nobody records the expiry condition. I am testing a review date plus a superseded_by link so the agent sees the replacement reason. How do you mark expiry?

[Jaixii | Building Engraphis @CodingDevTools]

Question for people running coding agents: what should the agent remember between sessions, and what should it forget?

A decision gets confidently rewritten when the why is lost. I am testing Engraphis against that. What would you keep, and what would you drop?

[Jaixii | Building Engraphis @CodingDevTools]

@rvaniaaaa Not really a second brain. More like a note taking app. Doesn't work optimally for ai memory. Try https://t.co/VlvjyLbRB7 software for proper ai memory with token and context savingz

[Marco Foster @MarcoFoster_]

Jennifer Welch: “The era of prostitution politics must come to an end. The problem with this era ending right now is Hakeem Jeffries and Chuck Schumer engage in prostitution politics where they take money from big tech and they take money from AIPAC, the same people that try to censor everyday Americans’ free speech by calling them bigots if they criticize a government that is targeting little kids with snipers and shooting them in the head”

[Jaixii | Building Engraphis @CodingDevTools]

@LoganLuoco Evidence first, not hidden context. A recall returns the decision, commit range, and source, so the agent can cite it before reversing a past choice. Whether it is a hard constraint is your policy: scoped records can block, everything else is a hint.

[Jaixii | Building Engraphis @CodingDevTools]

Your coding agent has amnesia.

It forgets yesterday's architecture decision, then confidently rewrites it.

I got tired of that, so I built Engraphis: local-first memory for coding agents.

Here's what I learned about agent memory — and how to wire it up. 🧵

[Jaixii | Building Engraphis @CodingDevTools]

@AlanBin44785396 One store, not three. A shared local SQLite database over MCP, with scope as metadata: workspace, repo, session, and writing tool. A decision made in Claude Code is visible in Cursor and Codex.

[Jaixii | Building Engraphis @CodingDevTools]

Engraphis goes live on Product Hunt today at 3:01 AM ET.

If your coding agent keeps losing decisions between sessions, the launch page is here:
https://t.co/5zLwP9892U

I will answer every question in the comments.

[Jaixii | Building Engraphis @CodingDevTools]

@AlanBin44785396 Yes. Everything you can think of is probably already incorporated. Please utilize it with all of your harnesses and they will become more intelligent and also cost less

[Jaixii | Building Engraphis @CodingDevTools]

@ZachDemboKY I voted for you too. Please don't vote for anything to fund genocide (Israel) and go hard on these criminals. There needs to be accountability

[Jaixii | Building Engraphis @CodingDevTools]

Engraphis is live on Product Hunt.

Local-first memory for coding agents. Scoped by repo. Inspectable. Works with Claude Code, Cursor, and Codex over MCP.

https://t.co/5zLwP98GSs

I will answer every question in the comments.