Agentic memory for your AI tools. A background daemon watches your platforms — code, conversations, commits, emails — synthesizes them into a living model of who you are, and serves it to every AI session automatically.

pipx install syke
syke setup --yes

That's it. Setup detects your data sources, ingests them, and starts the background daemon. Synthesis runs automatically on the daemon's first tick — no waiting.

uv tool install syke
syke setup --yes

git clone https://github.com/saxenauts/syke.git && cd syke
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
syke setup --yes

Syke supports multiple LLM providers. Setup shows a picker — choose whichever you have:

syke auth use codex             # ChatGPT Plus via Codex (reads ~/.codex/auth.json)
syke auth set openrouter --api-key YOUR_KEY  # OpenRouter
syke auth set zai --api-key YOUR_KEY         # z.ai

Claude Code session auth (claude login) is auto-detected if available, but is not the default — you pick your provider during setup.

Provider resolution: CLI --provider flag > SYKE_PROVIDER env var > ~/.syke/auth.json active_provider > auto-detect.

Switch providers: syke auth use codex or SYKE_PROVIDER=openrouter syke ask "question"

Check status: syke doctor shows active provider and credentials.

Every AI session starts from zero. Your context is scattered — commits in GitHub, threads in ChatGPT, sessions in Claude Code, emails in Gmail. Each tool sees a slice. None see you.

Syke fixes this. A background daemon syncs your platforms every 15 minutes, an AI agent synthesizes what it finds into memories, and the result — a self-evolving map of who you are, the memex — is distributed to your AI tools automatically.

The output — your memex:

# Memex — dev

## Identity
Full-stack engineer building AI developer tools. Python, TypeScript, React.

## What's Active
- **Syke v0.4.4** [high] (github, claude-code): Multi-provider support,
  Codex proxy, auth store, CI evolution. 261 tests.
- **Client project** [medium] (gmail, github): API integration due Friday.

## Context
Deep in release mode. Communication style: direct, fast-paced, technical.
Prefers concise answers. Working late nights this week.

## Recent Context
Shipping v0.4.4 multi-provider auth. Supports Claude Code, Codex,
OpenRouter, Zai. CI evolved with ruff linting. Next: observation layer.

---
Sources: claude-code, github, chatgpt, gmail. Events: 847.

This is what your AI tools see at the start of every session — instant context, no "tell me about yourself."

syke ask "question"   # Ask anything about yourself
syke record "note"    # Push an observation into memory
syke context          # Dump current memex to stdout
syke doctor           # Verify auth, daemon, DB health, provider status
syke auth use <name>  # Switch active provider (claude-login, codex, openrouter, zai)

That's the agent-facing API. ask spawns an AI agent that navigates your memories, follows links, cross-references platforms, and returns a grounded answer. context returns the memex instantly (local file read).

The loop: Collect signals → synthesize memories → update the map → serve to all tools → collect new signals from those tools → re-synthesize. Every 15 minutes.

Internally, the synthesis agent navigates your memory with 15 tools — full read/write access to create, link, update, supersede, and retire knowledge. It decides what's worth remembering. No heuristics, no embeddings — just an LLM reading and writing text.

syke daemon start     # Start background sync (every 15 min)
syke daemon stop      # Stop the daemon
syke daemon status    # Check if running, last sync time
syke sync             # Manual one-time sync
syke self-update      # Update to latest version

All data stays local in ~/.syke/data/{user}/syke.db — one SQLite file per user, copy it anywhere. Nothing leaves your machine except during synthesis (LLM API calls to your configured provider). A pre-collection content filter strips credentials and private messages before events enter the database.

Architecture — Four-layer memory system, synthesis loop, design decisions (why SQLite over vector DB, why free-form text, why Agent SDK)

Memex Evolution — 8-day replay showing how the memex graduates from status page to routing table

Setup Guide — Detailed installation, platform configuration, OAuth setup for Gmail

Docs Site — Full reference documentation

Demo — Live visualization of the synthesis process

Video walkthrough — 5-minute overview of how Syke works

MIT · Utkarsh Saxena