Your AI tools should
share a brain.
Claude doesn't know what you told ChatGPT. Copilot forgets what you decided last week. Sulci is the memory layer between you and every AI tool: it captures what you decide and serves it back, so you stop starting from zero.
sul·ci /ˈsʌlsaɪ/ · the folds and grooves of the cerebral cortex. Where memory lives.
Every session
starts from zero.
You use a handful of AI tools every day, and none of them share memory. A decision you made in ChatGPT is invisible to Claude. You re-explain your stack for the tenth time, re-litigate choices you already settled, and correct the same mistakes the model would not make if it remembered. The tool that knows your work is never the one you are talking to right now.
Your AI tools
SULCI
In context, everywhere
Context is captured once and served everywhere, automatically.
A memory layer
for every AI tool.
Sulci captures the knowledge from every AI conversation, the decisions you make, the preferences you set, the constraints you work under, and serves the relevant pieces back into whatever tool you reach for next. Not transcripts. Structured, ranked knowledge, injected before the model sees your first message.
Models are becoming commodities. Context is not.
In 2023, access to a frontier model was the advantage. Today that capability is available from a dozen providers at a few cents per million tokens, with open weights close behind. What does not commoditize is the layer above the model: the accumulated decisions, preferences, and institutional knowledge that make its output yours. That is where the durable advantage lives, and the platforms that rent you memory are quietly keeping it.
Context is not the same as information. Information is retrievable. Context is the difference between a colleague who joined last week and one who has worked beside you for two years: both have the same documentation, but only one knows why the schema looks the way it does, which decisions are sacred, and what feedback actually lands with you.
What changes
in practice.
You stop re-explaining yourself.
Sessions start in the middle of a conversation, not the beginning. The model already knows your stack, the decisions you made months ago, and the reasoning behind them.
Fewer correction loops.
When the model knows your preferences and constraints up front, it stops proposing things you would only reject. The correction loop is where most of the time in AI-assisted work actually goes.
Context that compounds.
Every session adds to a memory that is yours. Over months it becomes something a competitor who just signed up for the same API simply does not have.
Individually, each one saves time. Together they compound. For a team, the context logged over a year, the decisions, the conventions, the hard-won reasons behind them, becomes an asset a competitor on the same models simply cannot copy.
Structured atoms,
dual-indexed.
Every conversation is distilled into atoms: discrete, typed, scored units of knowledge, not raw transcripts. They live in a dual-indexed store, Postgres for structured queries and pgvector for semantic search, because each does what the other cannot. At the start of a session, Sulci embeds your query and injects only the atoms most relevant to what you are about to do.
Injected into every context response, regardless of query. For standing instructions and non-negotiable constraints.
Architectural choices already made. Gets retrieval priority when the query touches the decision domain.
Stack choices, implementation details, configuration. High weight on technical queries.
Style and behavior preferences. Ensures the AI works the way you work.
Temporal Validity
Atoms carry an expiry date. Expired atoms are excluded from all context responses automatically. Useful for sprint goals, code freezes, and time-limited constraints.
Atom Supersession
A new atom can supersede an old one. The old atom is marked expired on save. When a decision changes, the new one surfaces and the old one disappears. No cleanup needed.
The hard questions.
Written down.
Why atoms and not raw embeddings?
Raw embeddings preserve conversations but lose structure. An embedding of "we decided on PostgreSQL" sits semantically close to "we considered PostgreSQL." The atom model forces classification: one is a DECISION, one is noise. Classification changes what gets injected and when.
Why dual indexing?
A vector store finds semantically similar content but cannot answer "give me all CORE atoms" or "what expired this week." A relational store handles structured queries but cannot rank by semantic similarity. Both indexes run on the same data. Each query type hits the right one.
Why is Core Knowledge separate from retrieval?
Retrieval is query-dependent. An atom about your preferred test framework will not score high when you ask about deployment. But that preference should always be in context. CORE atoms bypass retrieval entirely. They go in first, every time.
Every way in.
One knowledge layer.
MCP Server
Claude Desktop & Claude Code
Native Model Context Protocol integration over HTTP (stdio for Claude Desktop). 17 tools spanning knowledge CRUD, context queries, conflict resolution, project management, and connected sources for Google and Slack.
Available Tools
Connected Sources
Google Docs · Calendar · Slack
OAuth integrations let Sulci read from and act across your workspace: search and read Google Docs, check your calendar, and search or post to Slack, all from the same context layer.
Available Tools
Import & Ingestion
GitHub · ChatGPT exports · Markdown · Obsidian
Bring existing knowledge in: index a GitHub repo, import ChatGPT memory exports, or drop in Markdown and Obsidian vaults. Share any project as a public, importable knowledge base.
Chrome Extension
ChatGPT, Claude, Gemini, Perplexity
Passive background capture. Conversations processed automatically. Nothing manual.
VS Code Extension
Visual Studio Code
Context sidebar in the editor. Relevant knowledge surfaces as you work, with a query box for your knowledge base.
API Proxy
OpenAI & Anthropic APIs
Transparent proxy that intercepts API calls, injects relevant context, and captures interactions. Zero code changes required.
REST API
Any Application
Full HTTP API for knowledge CRUD, context queries, interaction ingestion, data export, and privacy controls.
Dashboard
Visual Management
Browse knowledge, view entity graphs, inspect injection audit logs, manage exports.
Own it.
Don't rent it.
Most AI memory features store your context on someone else's infrastructure, in their format, under their terms. The context you build becomes their lock-in: switch providers and you start over. Sulci inverts that. What you accumulate is yours, inspectable and exportable, and it moves to any model you choose.
Privacy-First
Your knowledge stays on your machine by default. No cloud dependency. Export or purge at any time.
Provider Agnostic
Works across providers, tools, and workflows. Your context belongs to you, not to a platform.
Intelligent Decay
Relevance is a function of time, frequency, and confidence. Old decisions fade. Current context surfaces.
Fully Transparent
Every piece of injected context is logged and auditable. You can see exactly what Sulci told your AI, when, and why.
Teams & Multi-Tenant
Sulci Cloud runs on Supabase Auth with full tenant isolation. Spin up an organization, invite your team, switch between workspaces, and share projects, with pooled, seat-based billing.
“I explained my preferred stack to the same assistant for the tenth time in one project, always starting cold, losing decisions that had taken weeks to reach. The logs were all there. The reasoning was gone.”
So I built a context layer I own. Sulci started as a personal tool to fix that one frustration. It is now live at sulci.xyz, and every person who works across AI tools pays the same invisible tax I was paying.
Stop re-explaining yourself.
Two ways to get started. Self-host with full data control, or sign up and go in the cloud.
hello@lopez.fi
Let's build something
worth talking about.
I take on a limited number of advisory and fractional engagements. Only projects where I can make a real difference. If you're navigating growth, AI, or revenue challenges in a technical B2B environment, let's talk.