v0.1 — Public beta

Stop paying twice
for the same context.

StateSync is the global state database for AI coding agents. One shared memory across Cursor, Claude, OpenAI & Emergent — with 70% smaller context on every call.

Works withCursorClaudeOpenAIEmergentMCP
quickstart.py
# 2 lines. Any agent. Persistent memory. pip install agent-state-api from agent_state import StateSync sync = StateSync(api_key="ss_live_...") # Cursor saves the state sync.checkpoint( project="my-app", messages=history, files=["App.js", "server.py"], agent="cursor", ) # Claude Desktop resumes exactly where Cursor left off ctx = sync.pull(project="my-app", agent="claude") # → 12,400 tokens instead of 62,000 (80% smaller)
70–85%
avg. token reduction
5+
agents unified
<50ms
median sync latency
$0.005
saved per 1k input tokens
// features

Built for agents that actually ship code.

A middleware layer that gives every AI agent the same brain — with the cost profile of a cache and the reliability of a database.

Tiered State Storage

RAM + Disk architecture for agent memory. Hot messages stay verbatim, cold ones compress to dense semantic snapshots.

Multi-Agent Sync

Cursor, Claude, OpenAI, Emergent — one shared brain. Switch tools mid-project without losing a single decision.

Smart Context Compression

Hybrid semantic + temporal scoring keeps only mathematically relevant tokens. Verified 70–85% reduction on real projects.

Native MCP Support

Drop a JSON snippet into any MCP-compatible client. No plugins, no wrappers, no 50 different integrations.

Enterprise Audit Log

See which agent wrote which line, when, and why. Compliance-ready trails across your whole engineering org.

10-second Setup

`pip install`, paste your key, done. Works with the OpenAI Agents SDK and any framework speaking HTTP.

// how it works

Drop the MCP snippet.
That's the whole setup.

  1. 1Copy the JSON block into your MCP-compatible client config.
  2. 2Any checkpoint your agent creates is captured — files, decisions, tests, conversation.
  3. 3When you switch tools, the next agent gets a compressed brief instead of the full history.
  4. 4Watch your token bill drop and your projects stop breaking mid-refactor.
~/.cursor/mcp.json
{ "mcpServers": { "statesync": { "url": "https://api.statesync.dev/mcp", "headers": { "Authorization": "Bearer ss_live_..." } } } }
// pricing

Pay less than you save.

Every plan pays for itself after the first big context sync. Cancel anytime.

Free
$0/ month
  • 1,000 state syncs / month
  • +100 grace syncs
  • Up to 2 projects
  • 1 connected agent
  • Community support
Start free
Most popular
Pro
$29/ month per dev
or $290 / year (2 months free)
  • Unlimited state syncs
  • Unlimited projects
  • Up to 5 connected agents
  • Long-term memory across projects
  • Priority email support
Upgrade to Pro
Enterprise
$500+/ month
  • Everything in Pro
  • SSO + role-based access
  • Full audit log & compliance export
  • Dedicated context compression tuning
  • SLA + private Slack channel
Talk to sales
// faq

Answers, without the marketing fluff.

How does StateSync reduce token cost by 70%?+

Instead of resending your whole conversation, our compression engine scores each message by semantic relevance to the current task and temporal recency. Only the mathematically relevant slice is passed to the LLM, plus a dense summary of the rest. On real coding sessions we measure 70–85% token reduction.

Which agents are supported?+

Anything that speaks HTTP or MCP: Cursor, Claude Desktop, Windsurf, OpenAI Agents SDK, Emergent, custom Python/TypeScript agents. Two lines of SDK code or a JSON snippet in your MCP config.

Where is my code stored?+

Only conversation history, file paths, and decisions are stored — never file contents. Data is encrypted at rest, and Enterprise customers can request EU-region hosting or self-hosted deployment.

Is there a free tier?+

Yes. 1,000 free state syncs per month forever, no credit card required. Great for solo vibe coders.

Can I self-host?+

Enterprise plan includes on-prem / VPC deployment with dedicated support engineers.

Your agent forgets.
StateSync doesn't.

Cut your token bill by 70% and give every AI in your stack the same long-term memory.