One command to install. Works offline. Syncs when connected.

Memory isn't storage. It's a team practice.

When three developers use AI agents on the same codebase, those agents need to share what they know. Tages gives them persistent, quality-controlled, shared memory: conventions, decisions, architecture, past mistakes.

See how teams use it Try the demo

free MCP tools

<10ms

local recall

Any

model or machine

Terminal

# Add to any MCP client (Claude, Qwen, DeepSeek)

$ claude mcp add tages -- npx -y @tages/server

MCP server registered.

# Agent remembers conventions across sessions

$ tages recall "error handling"

[convention] api-errors

Return { error, code, status }

How it works

Add to Claude Code

MCP-only path needs no account. Or install the CLI and run tages init for cloud sync + dashboard.

claude mcp add tages -- npx -y @tages/server

# Or install the full CLI:
npm install -g @tages/cli && tages init

Remember

Store conventions, decisions, and architecture notes. Git hooks auto-extract from commits.

tages remember "auth-pattern" \
  "JWT in httpOnly cookie, refresh via /api/auth/refresh" \
  --type convention

Recall

Every new session starts with full context. No re-explanation needed.

# In any AI session, weeks later:
> "How does auth work in this project?"

# Agent recalls via MCP:
[convention] auth-pattern
  JWT in httpOnly cookie, refresh via /api/auth/refresh

AI agents have goldfish memory

Every session starts over. Your codebase knowledge vanishes between conversations.

Context amnesia costs hours

Every new AI session starts from scratch. The same questions get asked, the same codebase gets re-explored.

Naming conventions ignored

Your agent invented camelCase endpoints yesterday. Today it's writing snake_case. Last week it was kebab-case.

Same bugs reintroduced

You fixed that race condition last month. Your agent just recreated it because it never knew.

Three devs, zero shared context

Your agents work on the same codebase but make contradictory decisions. No one audits what they learn.

Why teams choose Tages

Not just another dev tool. Shared context that compounds.

Any model. Any machine.

Claude, Qwen, DeepSeek, Codex: one memory set works across every MCP-compatible tool. Switch models or machines without losing context. Cloud sync keeps memories in sync wherever you code.

Consistent conventions

Your agents follow the same naming patterns, error handling, and architecture decisions every time. No more re-explaining your codebase standards each session.

Local-first, sub-10ms

SQLite cache means your agent never waits. Works fully offline. Supabase syncs in the background when connected, so memories follow you across machines.