In development

Sinapt

A knowledge base for AI agents and machines.

Invisible. Persistent. Shareable. Your agents work normally — the context layer just persists for them across sessions, machines, and teammates.

Scroll

The shift

Agent-first, not human-first.

Your knowledge already exists. It lives in Slack, GitHub, meetings, docs, code, tickets — but it's queryable nowhere. Every agent rebuilds context from scratch, every session, every prompt. That's the knowledge tax. Sinapt is the layer that ends it.

Documentation tools

Notion. Confluence. Glean.

Wikis with AI bolted on. Optimized for humans browsing pages. Your agents are an afterthought — second-class citizens reaching into a UI that wasn't built for them.

Sinapt

Agent-first knowledge infrastructure.

Your agents query Sinapt directly via MCP, REST, or CLI. The context persists across sessions, machines, and teammates — invisibly. The web UI is for occasional human admin, not the daily surface.

What flows in

Stop re-explaining your context.

Your knowledge already lives in a dozen places — meetings, code, chat, docs, issues, files. Sinapt indexes it where it lives, into one layer your agents can query.

Built as a connector framework, not a fixed list. Add a source — all your agents see it.

Meetings & transcripts

Granola, Otter, Read.ai, Loom

Code & commits

GitHub, GitLab, local repos

Chat

Slack, Discord

Docs & wikis

Notion, Google Drive, Confluence, Obsidian

Tickets & PRs

Linear, Jira, GitHub Issues

Files & email

Local filesystem, S3, Gmail

And anything else with an API. Don't see your tool? Tell us →

Three surfaces, one backend

Built for agents. Available to humans.

Primary — for agents

MCP server

Your AI agents (Claude Code, Codex, anything MCP-aware) attach Sinapt as a standard knowledge source. No per-session re-context, no prompt-stuffing. Just persistent memory that's always there.

claude --mcp sinapt

Primary — for tools

REST API

All Sinapt operations exposed via clean HTTP. Custom scripts, automation pipelines, internal services, and CI all read from / write to the same backend your agents do.

POST /v1/query

Primary — for developers

sinapt CLI

Terminal-native client. Setup, ad-hoc queries, integration with your existing skills and git hooks. Bring your existing AI subscriptions — no token brokering, no AI markup.

$ sinapt query "..."

Secondary — for humans

Web UI

Clean app for occasional review, curation, team admin, billing. Not your daily surface — agents and CLI handle that. Web is for setup and oversight.

app.sinapt.ai

Roadmap

Two products, one platform.

Building now

Sinapt

The knowledge base.

Cloud-native infrastructure for AI agents. Markdown substrate, hybrid search, auth, teams, integrations, MCP server, REST API, CLI — everything an AI-native team needs to give their agents shared persistent context.

Roadmap

Sinapt Cockpit

One window, every agent.

Single-window desktop and mobile app. Drives Claude Code, Codex, and your existing CLIs. Layered on top of the Sinapt knowledge base. Ships after the KB has paying customers.

Security & privacy

Your data is yours. By design.

Security and privacy aren't features. They're the floor. For teams trusting Sinapt with knowledge they can't afford to lose, leak, or have read by anyone they didn't authorize.

Encryption by default

AES-256 at rest. TLS 1.3 in transit. Standard cloud-native primitives, not bolted on later.

Your data is portable

Plain markdown source format — export everything as .md files anytime. Open-source CLI under MIT license. Your tooling never depends on Sinapt's continued existence.

Never your training data

Customer knowledge base content is never used for model training. Not by us. Not by upstream model providers.

Real entity, real recourse

Backed by Petreski LLC — US legal entity (Wyoming, EIN 32-0778812). Two decades of engineering judgment behind every decision. Self-hostable enterprise tier on the roadmap.

Even our exit story is honest. Open-source CLI. Plain markdown source. Your data, your tooling, your control — independent of whether Sinapt exists tomorrow.

What ships, in order

Phases. Not deadlines.

Building in public. Each phase lands when it's ready — not when a slide said it would.

  1. Phase 1

    Architecture lock-in

    Infrastructure (AWS, IaC), search engine, storage, backend core, contract surfaces (CLI / MCP / REST). Hard problems first.

    Read: the architecture before the code →
  2. Phase 2

    Proof of concept

    Cloud knowledge layer running. CLI + MCP server live. Migrating real working setups as the test bed.

  3. Phase 3

    MVP

    Web app + team support + per-collection permissions + first integrations + billing + onboarding. The full product.

  4. Phase 4

    Public launch

    Open to teams. Named outreach. Sinapt Cockpit on the roadmap once Phase 3 has paying customers and a clear pull from operators.

Who's building this

Vanja Petreski.

Two decades in software engineering. AI-native operator since early 2025. Cockpit pattern across 10+ active repos. 1,282 documented hours of Claude Code use. Solo. Vendor-neutral by design.

Sinapt is two decades of engineering judgment applied to a problem the AI-native era created. Building in public — every architecture decision on the blog.

Follow the build

Be first to try.

Get a single email when there's something to try. No newsletter, no marketing — just the alpha drop.

No spam. Unsubscribe anytime.