Synapptic

synapptic analyzes your AI coding sessions and builds a living profile that your agent loads at the start of every conversation. Not just your preferences - it detects interaction patterns you didn't even notice: the corrections you keep making, the assumptions your AI gets wrong, the workflow quirks that matter to you but you never thought to write down.

The difference between memory files you write yourself (CLAUDE.md, .cursorrules) and what synapptic generates is that you can only document what you're aware of. synapptic sees the patterns underneath - the things that cause friction without you realizing why. It watches fifty sessions and tells your AI: "this person interrupts when you over-investigate, stops reading after the second paragraph, and will lose trust if you claim something without checking the code first."

The result is simple: you stop fighting the model. You stop repeating yourself. You get back into flow - the state where you're thinking about your code, not about how to make yourself understood.

Every synapptic install is personal. No two profiles are alike because no two developers are alike. Your profile reflects your communication style, your expertise, your frustrations, your standards. It's a fingerprint of how you work - built from your actual sessions, not from a template.

Get started

pip install synapptic
synapptic init       # pick your LLM provider and output targets
synapptic install    # set up automatic session processing
synapptic ingest     # analyze your existing sessions

That's it. From now on, every session ends with synapptic quietly learning in the background. The next session starts smarter.

What it builds

After analyzing your sessions, synapptic produces a living document with three sections:

## User Archetype

You are working with a senior full-stack engineer who expects execution,
not explanation. Terse commands, no pleasantries. Read diffs, don't
summarize them.

## Guards

1. NEVER commit without running tests first
2. BEFORE implementing a new service, read an existing one of the same type
3. WHEN the user specifies a verification path, treat it as a hard constraint
4. NEVER write a post-implementation summary

## Known Weaknesses

- Confident claims without evidence
- Scope creep on focused fixes
- Planning theater ("let me plan this" for implementation tasks)

This loads automatically at session start. Your AI already knows the rules before you type a word.

Works with everything

Any LLM provider for processing

Provider	Setup	Cost
Claude CLI	Already authenticated via Claude Code	Uses your plan
Anthropic API	API key	~$0.30-0.80/session
OpenAI API	API key	~$0.20-0.60/session
Ollama	Running locally	Free
LM Studio	Running locally	Free
Custom	Any OpenAI-compatible endpoint	Varies

Any AI coding assistant for output

Assistant	Where synapptic writes
Claude Code	~/.claude/projects/*/memory/user_archetype.md
Cursor	.cursor/rules/synapptic.mdc
GitHub Copilot	.github/copilot-instructions.md
Gemini Code Assist	.gemini/styleguide.md
OpenAI Codex CLI	AGENTS.md
Windsurf	.windsurfrules
Cline	.clinerules
Aider	CONVENTIONS.md
Continue.dev	.continuerules