Show HN: GreatArrow.ai – Shared memory for Claude, ChatGPT, Gemini and Cursor

I'll pull the product page and HN thread so the paragraphs stay grounded in what's actually stated, with no invented figures.Great Arrow is live as an open beta at greatarrow.ai and landed on Hacker News as a Show HN for shared memory across Claude, ChatGPT, Gemini, and Cursor. At post time the thread had 1 point and 0 comments. The product pitch is a single shared workspace that those assistants can read and write, plus the ability to take real actions in tools the user already runs, rather than starting every chat blank and pasting context by hand.

On the mechanics side, Great Arrow positions itself as a memory and tools layer over Model Context Protocol clients. It claims 17 first-class AI clients, 38 integrations such as Gmail, Slack, Drive, GitHub, Notion, and Jira, 315 actions those assistants can invoke, and 21 specialist agents. Memory is described as a four-layer model covering working, episodic, semantic, and procedural storage with embeddings for search; connected clients push conversations back over MCP, an auto-memorize step promotes important turns, and other clients search that store before answering. Security claims include AES-256-GCM at rest, Postgres row-level isolation per workspace, OAuth 2.1 with PKCE, hashed revocable API tokens, and prompt-injection screening on synced content. Cost controls break spend down per connected assistant and tool, with workspace budget ceilings and routing toward the cheapest capable model.

For engineers and builders, the useful claim is continuity across the tools they already pay for. An architecture decision made in Claude can be recalled later in Cursor; pricing terms found in Slack or Gmail can surface in ChatGPT without re-explaining the account. The product also tries to stop the human from being the integration layer by letting the same workspace send email, open pull requests, book meetings, or file tickets, with runs described as logged, costed, and reversible. Setup is framed as roughly five minutes, with a free trial tier and no credit card required, and install paths that include browser OAuth, one-click desktop installers, CLI, and .mcp.json.

Competitively, this sits between pure chat memory products and agent platforms that live in their own UI. Great Arrow stresses that it is not another tab: the same brain and actions follow the user across desktop, browser, IDE, and terminal, and switching AI vendors does not wipe history. That puts it next to vendor-native memory features inside Claude, ChatGPT, Gemini, and Cursor, and next to other MCP tool hubs, but with a sharper cross-vendor continuity and action story. Pricing is published as Personal at $39 per month for about 70 grounded conversations with memory and chat only, Pro at $119 per month for about 180 grounded conversations or about 30 multi-step agent runs with full orchestration and daily sync, and custom Enterprise with multi-seat roles, token caps, and bring-your-own encryption key. Plans include audit logging, bank-grade encryption, secure OAuth, and the 315 built-in actions, with agent actions on Pro and above; each plan covers underlying AI API spend up to a monthly token cap, with optional capped pay-as-you-go overage.

The practical next step is whether the open beta proves the shared-memory loop under real multi-client use without context rot or unsafe tool calls. Watch for how reliably decisions written in one client reappear in another with sources attached, how reversible agent runs feel in day-to-day workflows, and whether cost attribution per surface stays predictable once several AIs share the same workspace. The product site and sign-up flow are at https://www.greatarrow.ai; the Show HN thread is https://news.ycombinator.com/item?id=49203752.