Getting Started

Connect your editor.Memory works instantly.

Sign up, pick your IDE, and your AI assistant gains persistent memory across every session.

01

Sign up

Create a free account. An API key is generated automatically.

02

Connect your IDE

One click, one command, or a JSON snippet — pick your editor below.

03

Start coding

Your assistant saves and recalls context across sessions. No prompts to remember.

Connect your IDE

11 editors and CLIs supported. Pick yours below.

One-Click InstallClick a button in your dashboard — the IDE opens with a confirmation prompt.

CLI CommandPaste a single command into your terminal.

Manual ConfigCopy the JSON snippet into your IDE's config file.

Your first session

Open your IDE and start a conversation. Everything else is automatic.

01

The agent calls session_bootstrap

On the first message in any project, your coding assistant detects Mnemonik and initializes. It registers the project from the working directory and loads any existing context.

02

Three files are written to your project

A .mnemonik.json identity file (stable ID across renames and team members), an IDE rule file, and a skill file. The rule tells the agent "you have Mnemonik, use it." The skill teaches the full workflow. You never write these manually — they're updated automatically when Mnemonik ships new versions.

03

Memory tools work immediately

memory_add, memory_search, file_context, checkpoint — all active from the first message. Every coding session carries forward decisions, patterns, and context.

Code context & the scanner daemon

Memory tools work instantly, but code-level context (file_context) needs your files indexed. Mnemonik handles this automatically with two mechanisms.

Scanner daemon starts automatically

On first session_bootstrap, the agent detects whether the daemon is running. If not, it spawns it in the background via npx @mnemonik/scanner. The daemon fetches your project list from the server, scans every project on your machine, and watches for file changes. One process covers all projects. You never interact with it.

On-demand fallback fills gaps

If the daemon hasn't indexed a file yet (e.g., just created), the coding assistant reads it locally and pushes the content to the server via file_push. This fills gaps transparently — you never notice.

Anti-drift proxy

Claude Code only — optional but recommended

MCP gives the coding agent its memory tools. The anti-drift proxy is a separate layer that keeps it using them throughout long sessions — not just at the start.

Session reinforcement

A system-reminder is injected into every API call, reinforcing Mnemonik tool usage even after context compression.

Auto-distill

When a conversation reaches 20+ messages, the proxy automatically extracts structured memories from the transcript. Non-blocking, with a 2-minute cooldown.

Get a proxy token (pxt_...) from your dashboard. This is separate from your MCP API key.

terminal
claude config set --global apiBaseUrl "https://api.mnemonik.dev/proxy/pxt_YOUR_PROXY_TOKEN"

Teams

Team plans share a project identity across members while keeping each developer's context personal.

Shared project, personal context

All team members use the same .mnemonik.json and project ID. At query time, file_context prefers your version of files you've modified — you see your uncommitted changes, not a teammate's. For files you haven't touched, you see the team's collective knowledge.

One daemon per member

Each member's scanner daemon runs independently. Content hash deduplication at the server prevents redundant embedding costs. The shared knowledge base is an emergent property — no CI infrastructure, no git webhooks, no shared service accounts.

New members get context immediately

When a new team member joins, they get code context from existing scans right away. No onboarding lag.

Ready to try it?

Sign Up Free