AI capabilities
Jingle status for AI commands, agent tools, context, approvals, and inspectable results.
Jingle's AI direction is not another chat box. AI should use extension tools, local context, and explicitly configured user preferences within boundaries that are visible, approvable, and replayable by the user.
Capability Status
| Capability | Status | Target |
|---|---|---|
| AI Chat | Preview | Work with the current workspace, local context, and extension tools |
| Quick AI | In development | Start lightweight AI tasks from the current context without interrupting the main workflow |
| AI Commands | In development | Turn repeated prompts into reusable commands with input, model, and output contracts |
| Agent tool calls | Preview | Extensions expose tools that agents call within user-visible boundaries |
| Context selection | In development | Combine files, memory, connection state, selected text, and command input for a task |
| AI Profile | Planned | User-editable preferences, writing style, common context, and personalization boundaries |
| Human approval | Preview | Show tool name, arguments, description, and risk before execution |
| Schema-form approval | Planned | Let users edit arguments, confirm a form, and then execute the tool |
| Result explanation | In development | Present tool results in a human-readable way and provide recovery paths on failure |
| Evals | Planned | Cover use cases, permissions, arguments, and failure semantics with automated checks |
Platform Boundary
- Agents should not directly receive access tokens, refresh tokens, or client secrets.
- Agents can see tool descriptions, connection state, input arguments, and approval outcomes.
- AI profile / personalization belongs in user-editable settings, not invisible profiling.
- Extension tools execute in the main service, while the host owns permission, connection resolution, and result presentation.
- Tools that mutate external systems need richer approval and argument display than a simple confirmation button.
Product Direction
- Stabilize tool registration, connection state, and presentation metadata.
- Add schema-form approval and argument editing for mutating tools.
- Add end-to-end evals for provider tools such as GitHub and Notion.
- Turn AI Commands from prompts into reusable, publishable, and shareable command units.
- Make Quick AI and AI Profile visible, editable, and optional in the product.