Stop re-explaining yourself

Cognograph is a spatial AI canvas where conversations, agents, terminals, notes, tasks, and automations live on one canvas. You connect nodes with edges and the AI reads the graph, assembling context via breadth-first traversal instead of forcing you to copy-paste between windows. 9 node types, 13 spatial trigger types, open source AGPL-3.0.

A workspace where layout is context engineering. Instead of managing context through system prompts and config files, you arrange nodes on a canvas and draw connections between them.

The AI traverses those connections to build its context window. Your spatial layout becomes your prompt engineering.

I built Cognograph because I have ASD and severe combined-type ADHD. Linear chat doesn't match how associative thinkers process information. Spatial layout maps to working memory: persistent, visible, connected. It started as a workaround for my brain. Turns out spatial thinking isn't a neurodivergent workaround. It's how most people handle complexity when given the option.

Bring your own API key for Anthropic (Claude), OpenAI, Google (Gemini), or OpenRouter (100+ models). Or skip API costs entirely: connect Ollama for local inference with Llama, Mistral, Phi, or any GGUF model. Terminal nodes support any CLI-based agent. Each conversation node can use a different provider.

The spatial graph walks connections and sends what's linked. Fewer tokens per call, any provider.

When you connect nodes on the canvas, Cognograph walks the graph via breadth-first traversal and assembles only the relevant context for each AI call. Connected nodes become the AI's working memory. You send what matters, not everything. In my testing, this cuts token usage up to 80% compared to re-prompting the same context every message.

Spatial triggers are patent-pending automation mechanisms that fire based on canvas state. 13 trigger types: proximity detection, region enter/exit, cron schedules, property changes, connection events, cluster size thresholds, child completion, ancestor changes, connection count, isolation detection, manual, and node creation.

Your canvas layout becomes executable logic. No scripting required.

Semantic zoom handles scale. Five levels adapt automatically: ultra-far shows cluster shapes, far shows titles, mid shows lede text, close shows full content, ultra-close expands toolbars and property panels. The graph also self-organizes relevance through BFS traversal. You connect what matters and the AI curates what's relevant.

Yes. The Electron desktop app runs with zero server dependency. Connect Ollama or any local model for fully air-gapped AI. Workspaces save as local JSON files. API keys are encrypted via your OS keychain, never sent to Cognograph servers.

Browser: IndexedDB, never leaves your machine. Desktop: local JSON files on your filesystem. Pro tier: cloud sync with end-to-end encryption. No telemetry in any tier. No account required for free use.

Chat is one thread. Context dies when the window closes. A spatial canvas is persistent nodes you wire together. The AI reads the graph topology, not just the last few messages. Context engineering replaces copy-paste. You build a knowledge graph that gets smarter as you connect it.

Workflow builders run linear pipelines: if this, then that. Cognograph is a graph where edges carry context in any direction. Nodes aren't steps in a sequence. They're persistent context the AI traverses. The canvas is the programming language.

AGPL-3.0. 900+ source files on GitHub. 2,400+ tests across 141 test files. Read every line, run it yourself. Self-hostable on any machine with Node.js 20+. Free for personal use. Commercial use requires a commercial license.

Four provisional patent families covering how Cognograph assembles context, orchestrates spatial layouts, fires triggers, and previews AI-proposed changes. The Defensive Patent Pledge means I won't use these patents against anyone unless they come after Cognograph, its users, or contributors first. The patents protect the community, not restrict it.

Three options. Browser: $0, bring your own API keys, full canvas, no limits. Desktop: $0, self-host, bring your own keys or run Ollama for zero-cost local AI. Pro: $15/mo, cloud sync with E2E encryption, $5 AI credit included, works out of the box.

Browser: go to cognograph.app/workspace. No account required. Desktop: git clone https://github.com/skovalik/cognograph.git, npm install, npm run dev. Requires Node.js 20+.