feat(client): add kai-agent backend support — new endpoints and approval flow

[Matovidlo]

Release Notes

• Adds submit_approval() method using the new dedicated POST /api/chat/{id}/approval endpoint — replaces the complex Vercel AI SDK v6 approval protocol
• Adds get_usage() to retrieve rate-limit usage (messages used, limit, reset date)
• Adds get_settings() / update_settings() for project-level custom instructions
• Adds get_user_settings() / update_user_settings() for user-level custom instructions and tool permissions
• Adds get_tools() to list available MCP tools with read-only flags
• Adds get_suggestions() for AI-generated contextual suggestions (dashboard, job-detail, configuration-detail)
• Deprecates old v6 approval methods (approve_tool, reject_tool, send_tool_approval_response, send_tool_result, confirm_tool, deny_tool, resume_stream) — they are incompatible with the new backend

Plans for customer communication

None.

Impact analysis

• KaiClient: 8 new public methods added — fully backwards-compatible, no existing API changed
• models.py: 8 new response models (ApprovalResponse, UsageResponse, AgentSettings, UserAgentSettings, ToolInfo, ToolsListResponse, Suggestion, SuggestionsResponse) — all exported from __init__.py
• Approval flow change: the new submit_approval() is a non-streaming POST that unblocks the live SSE stream still connected from send_message(), matching the kai-agent backend's in-memory approval store design
• Deprecated methods still exist and still compile — no breaking changes for existing callers
• 28 new tests added, all 89 tests pass

Change type

Feature — Connect kai-client to the new kai-agent backend

Justification

The kai-agent backend (keboola/ui apps/kai-agent) is a new Hono-based server replacing the old Vercel AI SDK v6 backend. It exposes several new endpoint groups (usage, settings, suggestions) and uses a fundamentally different tool approval mechanism: instead of re-POSTing a mutated assistant message, callers now submit approval decisions via a dedicated side-channel endpoint while the original SSE stream stays connected and resumes automatically.

Deployment

Merge & automatic deploy.

Rollback plan

Revert of this PR.

Post release support plan

None.

[Copilot]

Pull request overview

Adds kai-agent backend support to kai-client by introducing new client methods, response models, and tests for the new endpoint groups (approval side-channel, usage, settings, tools, suggestions), while marking legacy Vercel AI SDK v6 approval helpers as deprecated in docstrings.

Changes:

• Added KaiClient.submit_approval() for the new non-streaming POST /api/chat/{id}/approval flow, plus new client methods for usage, settings, tools, and suggestions endpoints.
• Introduced new Pydantic response models for the added endpoints and exported them from kai_client.
• Added a comprehensive set of tests covering the new endpoints and request payloads/headers.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
src/kai_client/client.py	Adds new public client methods for kai-agent endpoints and updates legacy approval method docstrings with deprecation notices.
src/kai_client/models.py	Adds new response/settings/tool/suggestion models to parse kai-agent backend responses.
src/kai_client/__init__.py	Exports the new models from the package public API.
tests/test_client.py	Adds tests validating new endpoints, payload serialization, auth headers, and error handling.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.