Orbit

Orbit, built by Fraima, is a shared memory and coordination layer for people running work across several AI tools at once. Rather than being an agent that resolves a single task, it sits on top of Claude, ChatGPT, Cursor, Codex, Gemini, Manus and n8n and gives them one workspace: a persistent memory every session reads before it acts, a live dashboard of what each agent is doing, and a searchable vault where every output is filed automatically. It is aimed at solopreneurs, freelancers and small teams who have AI producing work in a dozen places with no single source of truth for what was decided or done. Integration is the strongest part of the product. Orbit ships its own Model Context Protocol server: you paste one MCP command into Claude or another MCP client and it connects in about two minutes, with no OAuth or webhooks. Through that endpoint external clients read Orbit memory, vault files and agent outputs, and Orbit agents can call outward to tools such as GitHub, Notion, Linear, Stripe and Gmail, so the MCP layer runs in both directions. The server is real and first-party, and Orbit is independently listed on public MCP directories. Named coverage today centers on Claude, ChatGPT and Codex, Cursor, Gemini, Manus and n8n, with deeper native integrations beyond MCP still being added. Pricing is a genuine freemium ladder billed monthly and listed in US dollars. The free plan is permanent and includes 100 credits a month, one Orbit and seven days of history. Pro is $8 per month with 500 credits, unlimited projects and full vault history; Premium is $24 with 2,000 credits, up to five Orbits, scheduling and insights; Team is $39 with 5,000 credits, 25 Orbits, five seats and an audit log; and Enterprise is quote-only. Every plan is metered in credits, and Orbit actions consume roughly double the credits of a read, so the practical cost of running several agents continuously sits above the sticker price. What the entry price does not buy is unlimited autonomous operation. The credit cap is the real constraint: heavy multi-agent use will exhaust a monthly allotment, and the fully autonomous runtime, Agent Builder and Autonomous Mode are early-access rather than the finished autonomous AI company the marketing describes. Today the dependable value is shared memory, observability and coordination with a human approving what becomes canonical, not hands-off delegation. Orbit is not the right fit for teams that need a proven, audited platform. There is no SOC 2 or ISO certification yet; security rests on row-level isolation, TLS and EU data residency with full GDPR rights. There is also no independent review base on G2, Capterra or Product Hunt, so buyers cannot yet gauge sentiment at scale. On data handling it is stronger than most: Orbit states it does not use workspace content to train foundation models, and its model providers do not train on it under their terms. As of Q3 2026 it is a credible, early-stage entrant whose differentiator is a real bidirectional MCP layer aimed squarely at heavy Claude and Cowork users.