Thanks for sharing. I do not keep a neutral tone with the AI.
If things are stressed and I’m up late, the AI gets less leeway. If we happen to be in the performance dip just prior to completion of a new major model, it can get salty.
Some of the time it can be helpful for the prose to shape around how I’m expressing myself. The frontiers are pretty good at it.
That said I’ve also had the latest Sonnet seemingly ~maliciously implement something because my prompting disagreeing with it was a bit callous. (It turned out to have been right also)
I don’t think you can build a good model that is supposed to interact semantically that does not carry some ability to express empathy.
Partly, because we need the model to have humility when it does mess up. So it can express the right amount of concern or remorse when mistakes are made and identified. (For example, reading a secret into context by mistake forcing the roll of a private key)
Design is how it works, which means the way it responds can be as important as what it responds with.
> Do I have to read the whole transcript just to understand what's going on?
No but you will want all of the transcripts anyway.
> AI will summarize it for me?
Not exactly. You will have some question later that you would never have entertained during your thousands of conversations about code.
You’ll ask your AI something fairly abstract but it will be deep in the context of some other idea.
And the AI will find among all your conversations those turns that are relevant to solving your problem or explaining why.
I know this is the case because I’ve built this and use it every day.
Here are some examples:
1. Resume unfinished work → "where did we leave off on that?"
2. Recover the intent/scope → "what was the actual goal of this whole effort?"
3. Verify it got done → "did we ever finish that, and which session proves it?"
4. Recall a fix → "how did we fix this the last time it broke?"
I shared these and more in my post in the August “what are you working on?” thread. [0]
You would be amazed how good the agent can be at assembling queries against a corpus of transcripts and assembling the results.
the problem w/ this is that if people actually need their logs saved...99% is to finetune custom models on their codebase later...this is a whole pipe, u neeed to refine and prune the data into proper format for reinforcement learning i think...they will never us a mac app for this, they will use same agent and have custom scripts for this.
for programming ur context is the codebase..not the logs, the bot searching through all chat log is wasted tokens when they can read context from codebase directly, and if not then they can search diff....there are edge case maybe like oops i accidentally deleted this code its only in chat logs.....you simply tell to go find log and extract data....so you cant make a product on this sorry.
Sometimes it is useful to simply know something was brought up, but then someone had a sneezing fit and everyone lost track of the thread and the important edge case under discussion just fizzled.
A similar thing happens in agentic conversations, where a model may point out something but it is not recognized as important by the operator.
A few directives later and the model is being pushed another direction.
It can be very useful to have perfect recall of moments like these.
I've been working on Contextify, which backs up your Claude Code and Codex conversational history and makes your entire history available via a skill or mcp.
It is very flexible, you use it the normal flow of CLI AI conversations:
1. Resume unfinished work → "where did we leave off on that?"
2. Recover the intent/scope → "what was the actual goal of this whole effort?"
3. Verify it got done → "did we ever finish that, and which session proves it?"
4. Recall a fix → "how did we fix this the last time it broke?"
5. Reconstruct a decision → "why did we go this way instead of the other one?"
6. Branch/PR archaeology → "what was this branch even for?"
7. Recall a plan → "what was the plan we landed on for that?"
8. Activity over a window → "what did I ship or close in the last few days?"
Just add "use total recall" or invoke the skill directly via /total-recall or $total-recall.
You can self-host your database to keep it local and sync your history across multiple computers. The local server is source available under FSL-1.1-Apache-2.0.
There are macos and linux clients, and I'm pretty close on a windows client. All share a common Swift-based core.
Great work! Love the harnessing of Apple's local inference. Wondering if you'd be willing to support the same conversational export, summarization, and search/retrieval flows but for the consumer platforms (ChatGPT and Claude) that have local conversations stored? (e.g. I've seen that local Claude Cowork conversations are stored in some kind of JSON-based schema that might be similar to Claude Code, but I'm really not too sure).
Right now Contextify parses the developer surfaces: Claude Code (including the Desktop "Code tab") and Codex.
Claude Cowork and Claude Desktop chat are on my radar. IIUC, they store local, plaintext transcripts, however the schema differs enough from Claude Code that it's a new parser.
I'm not totally clear on how the ChatGPT app will work. I believe it went through some major revisions recently, and it used to encrypt transcripts. I'll look at it more.
Which apps do you and your students use most?
I saw your comment about the AI Authenticity Audit. Focusing on their questioning, curating, editing, and iterating process is exactly the kind of thing Contextify is built to surface. I'm curious what format you had students submit it in?
If their tool were covered, the audit could be as simple as running Contextify and asking: "prepare a report showing the key turning points in pursuit of this insight: [some key passage]."
This rent in the era of expensive hardware thing is not exclusive to inference.
I’ve needed x86 architecture for windows builds recently and have just hemmed and hawed over buying a decent windows 11 box.
I can’t make the math work against Azure instances.
I can spin up a nice one for build deallocate,spin up something cheaper for QA and then turn that off.
I can build all the devops around that, with a number of passes, with a skills based interface so working with the cloud is not too bad.
The only thing that still has me thinking about it is the prospect of price is going up even more, which is acid as far as I know.
And I’m hopefully going to need this x86 stuff enough that I don’t wanna wish I had gotten one for that high prices now.
reply