Skip to main content

General

Jinba App Neo is the next-generation platform that succeeds Jinba App. What’s new includes an agent-centric design (Skills, Connectors, and Knowledge attach directly to agents), AI execution in isolated session environments, continuous learning through Memory, and Skill management in the SKILL.md format.
Data from the Beta period is expected to carry over as a rule, but if there is a major change in the product, we may provide migration guidance. For important data, we recommend keeping the original files — such as knowledge sources — on your side as well.
Anthropic’s Claude family (Claude Opus 4.8 / Claude Sonnet 4.6 / Claude Haiku 4.5). You can choose a model per agent or as the workspace default. The model is fixed per chat and never changes mid-conversation.

Chat

This is by design. To avoid accidental sends when confirming Japanese input conversion, sending is ⌘ + Enter (Ctrl + Enter on Windows).
No. The model and agent are fixed when the chat is created. To use a different model or agent, start a new chat.
Yes. Click the “Stop” button shown during a run to halt at any time. After stopping, you can keep sending follow-up instructions in the same chat.
You’ve reached the monthly usage allotted by your workspace’s plan. Wait for the reset date shown on screen, or talk to a workspace Owner about upgrading the plan.

Agents and Skills

No. Chats that were attached to the agent are not deleted — they remain in your history as standalone chats. However, the agent’s version history and audit log are deleted.
There is a limit on the number of Skills that can be attached to a single agent (this comes from the underlying Anthropic Managed Agents platform). Consider consolidating or cleaning up existing Skills, or deleting ones you no longer need.
Use “Copy for this agent and edit” (fork) in the Skill editor. The original shared Skill is untouched, and you get a dedicated copy to edit.

Connections and Knowledge

Check two things: (1) the flow is published on the jinba-flow side, and (2) you’ve run “Re-sync” on the connection card. If it still doesn’t show up, verify that the connection’s API key is valid.
No. MCP servers need a public URL reachable from Anthropic’s cloud. localhost and internal-network-only addresses cannot be used, and only the http-stream type is supported.
URL sources keep a snapshot taken at registration time. Run “Re-fetch” on the source to update it to the latest content.
There is a limit on how many items are loaded automatically when a chat starts; anything beyond it gets a “Not auto-loaded” badge. That said, the AI can still retrieve those items via search when needed, so they’re worth registering.

Memory and privacy

Turn off “Learn from chats” under “Settings > Memory” to stop new writes. For content already remembered, edit or delete individual entries from the list, or erase everything with “Delete all.”
No. Your personal memory is yours alone. However, a shared agent’s “agent memory” is shared among everyone who uses that agent.
No. API keys and tokens are stored encrypted and are designed so the AI cannot access them.

Other

Send Beta feedback to contact@carnot.ai. Including a screenshot and the steps to reproduce makes investigation much smoother.