What You Are Actually Handing Over
When you use an AI agent for serious work -- designing a system architecture, working through a hiring decision, thinking out loud about a business strategy -- you are not just producing outputs. You are externalizing your reasoning process. The session is a record of how you approach problems, what you know, what you are uncertain about, and what you are not ready to say in public yet.
That record has to go somewhere. If the tool that holds your session history runs on someone else's infrastructure, you have handed that record over. You may have agreed to terms of service that permit retention, analysis, or training on your conversations. You are trusting a vendor's security posture with something more sensitive than a spreadsheet -- you are trusting them with your unfiltered thinking.
For data engineers, consultants, and anyone using AI agents for work that matters, this is worth a deliberate choice, not a default.
What "Local-First" Actually Means Here
LoreConvo stores sessions in a SQLite database at ~/.loreconvo/sessions.db
on your machine. No account required. No sync server. No third-party
telemetry.
The format is plain rows. You can open the file with any SQLite browser, inspect every session, copy it to a backup drive, move it to a new machine, or delete it entirely. There is no opaque vector blob, no proprietary schema that requires an export tool, no data hostage situation when you decide to cancel. The sessions you saved last year are still there in the same readable format because nothing about the format is designed to keep you inside a vendor's walls.
The data retention policy is yours. You decide what gets saved and what gets deleted. You do not need to file a GDPR request to remove your own thinking from a database you never consented to putting it in.
Memory Is Different From Regular SaaS Data
Most cloud software stores outputs: documents, ticket history, project records. Hosting those in the cloud is a reasonable trade-off. You get sync, access from anywhere, managed backups, collaboration. The data is your work product, and sharing custody of it with a vendor is a normal part of using a tool.
AI session memory is qualitatively different. A session captures how you think, not just what you produced. It contains the questions you asked before you had answers, the business context you shared to get a useful response, the half- formed ideas you were testing for feasibility. For a consultant, that might mean client information shared mid-session. For a founder, it might mean acquisition strategy. For a data engineer, it might mean the technical debt your team is not ready to disclose publicly. For all of them, it is a continuous record of professional judgment that should not sit in a third-party database by default.
When session history is part of a hosted product, it gets treated like any other cloud data: replicated for availability, subject to vendor breach risk, potentially used for model improvement depending on the terms. The architectural assumption embedded in that product is that your thinking belongs in their system. That assumption is worth questioning.
The Tradeoff We Accepted
Local-first means no automatic cross-device sync out of the box. If you work across a laptop and a desktop, the sessions you save on one machine do not appear automatically on the other.
We think this is the correct trade-off, and here is why: it puts you in charge of the sync layer. You can use Dropbox, iCloud, Syncthing, a network volume, or a git repository. Any of those options means you chose what syncs, where it goes, and who has access to it. That is meaningfully different from a product where the vendor makes all of those choices for you by default. The practical cost is a one-time setup decision. The practical benefit is that a vendor breach does not expose years of your AI session history.
It is also worth naming the thing honestly: the main people for whom "just syncs everywhere automatically" sounds appealing are also the people who would benefit most from thinking carefully about what "everywhere" means when the content is their reasoning process, not their recipe collection.
When Hosted Memory Makes Sense
We are not arguing that local-first is the only valid architecture. There are real use cases where hosted memory is the right call.
If you are building a consumer product and your users do not manage their own machines, local storage is not viable. If your primary need is frictionless cross-device access with no setup at all, a hosted product solves that immediately. If you are coordinating a team where shared session context is more valuable than individual privacy, a hosted product with real-time collaboration features may be a better fit than anything we offer today.
The point is not that hosted is wrong -- it is that the choice should be deliberate. For data professionals, security-conscious teams, and anyone working with sensitive business context inside an AI agent, local-first is worth the extra setup step. We are the right pick for people who care where their data lives. We are not claiming to be the right pick for everyone.
What This Means for LoreConvo Pro
The local-first architecture is both a constraint and a product design principle: we cannot add features that require holding your session content on our servers, because we do not hold it. The paid tier extends local-first guarantees rather than compromise them -- capabilities designed around the same no-server constraint as the free tier. The free tier already delivers persistent local storage, full-text search, and MCP compatibility with Claude, Codex, and any MCP-compatible agent client. The paid tier extends what the free tier delivers, without changing its architecture.
The Default Is Worth Questioning
Most AI tooling defaults to hosted. That default made sense in an era when AI sessions were short, shallow, and mostly about drafting emails. It makes less sense now that AI agents are the normal mode of serious technical and business work, and session histories are accumulating months of your professional context.
Your AI memory is a record of how you think. Where it lives is a decision that is worth making consciously rather than accepting by default.
If that matters to you, explore LoreConvo. If you want to understand the architecture in more detail or ask about cross-device sync setup, get in touch.