fa805a28a41aca0c1903beba3a7ea44fc675d8dc
10 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
| 067574ad6a |
feat(ai): robust response extraction + structured-output foundation
Harden the Anthropic provider and lay the groundwork for schema-constrained JSON, optimizing the existing claude-sonnet-4-6 / claude-haiku-4-5 usage (no model changes). ai_provider.py: - _extract_text_from_response replaces fragile response.content[0].text: skips non-text leading blocks (e.g. thinking), returns the first text block, logs an anthropic.stop_reason warning on max_tokens/refusal (truncation now observable), and raises ValueError on a no-text response. - generate_json gains an optional `schema` param. Anthropic wires it to output_config.format (structured outputs); schema=None preserves the exact prior call for every existing caller. Gemini accepts-and-ignores it. kb_conversion_service.py: - TROUBLESHOOTING_SCHEMA / PROCEDURAL_SCHEMA + _schema_for_target_type(), modelled as a strict superset of every field the prompts emit. - convert_document passes the schema only when the new AI_KB_CONVERT_STRUCTURED_OUTPUT setting is True (default False). The _try_repair_json fallback stays as belt-and-suspenders. Tests: 14 provider + 7 schema, TDD (red-green). Live constrained-decoding smoke-test still required before enabling the flag in production. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> |
|||
| 3f0a132058 |
refactor(ai): rename _call_anthropic_cached → chat_call_cached; extract cache plumbing (Phase 0.4)
Renames the chat caller to a name that signals its actual purpose, and factors the reusable cached-system-block + cached-history + cache-usage-log primitives out to app.core.ai_provider so they can be shared with the provider-generic path without pulling MCP/beta/images into the abstract interface. Helpers added to ai_provider.py: - `build_anthropic_chat_messages(history, new_message, images, format_reminder)` — owns: copy history, apply cache_control to last history message, append format reminder to new message, render images as multimodal blocks. Anthropic-shaped by design; do not call from Gemini paths. chat_call_cached keeps exactly the concerns that are unique to the one MCP/beta/multimodal chat caller: - Anthropic beta endpoint invocation - Microsoft Learn MCP server wiring (ENABLE_MCP_MICROSOFT_LEARN) - Retry-without-MCP fallback - Format-reminder content string (declared as module constant) - Phase 0.5 telemetry (mcp.turn, mcp.fallback) Documents in the module docstring AND at the function site that this is the ONE MCP/beta chat caller and should not become the general provider path. MCP/beta/images are features of exactly one optional Anthropic beta endpoint; routing them through AnthropicProvider would leak a provider- specific concern into the abstract interface that also serves Gemini. Behavior change: chat_call_cached now reuses the singleton AnthropicProvider HTTP client via `_get_anthropic_client(...)` instead of instantiating a new `anthropic.AsyncAnthropic(...)` per call. Matches the provider's own pattern and avoids burning connections per-turn. No user-visible difference. No runtime verification from code-server. TODO(phase0-verify) in ai_provider.py tracks the cache-hit verification owed on the new dev env. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
|||
| b3be66652e |
feat(ai): structured-system-block caching in AnthropicProvider (Phase 0.1)
Widens AIProvider.generate_json / generate_text / generate_text_stream
signatures to accept `system_prompt: str | list[SystemBlock]`:
- `str` (the existing call shape): passes through uncached, unchanged
behavior. Every existing caller stays on the uncached path — no silent
behavior change.
- `list[SystemBlock]`: enables Anthropic prompt caching via structured
system blocks. Caller-authored `cache_control` is honored verbatim
(policy α); if no block carries it, the provider applies
`cache_control: {"type": "ephemeral"}` to the first block only.
Gemini ignores cache_control and concatenates list entries into one
system string — the widened signature is strictly additive on that path.
Adds `anthropic.cache` structured-log telemetry: on every Anthropic
response (streaming included, via `stream.get_final_message()`), logs
`cache_read_input_tokens` and `cache_creation_input_tokens`. Telemetry
failure in streaming is swallowed so the user-facing stream never breaks.
Verification deferred: cannot run from code-server (no Python, no DB,
no dev env). TODO(phase0-verify) left inline in the module docstring.
First verification task on the new dev environment is to hit any
FlowPilot endpoint twice within 5 minutes and confirm the second call
shows cache_read_input_tokens > 0 in the `anthropic.cache` log event.
If verification fails, that's a debug task on the new env — not a
blocker for continuing Phase 0.2/0.3/0.4.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|||
chihlasm
|
2f3781bfc2 |
feat: add generate_text_stream to AnthropicProvider for SSE support
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
chihlasm
|
c6772c6607 |
perf: singleton AsyncAnthropic client to avoid per-call connection setup
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
chihlasm
|
0dc6123c0c |
feat: flow export/import + procedural Flow Assist (#96)
* feat: add flow export/import backend (migration, endpoints, schemas)
Add .rfflow file export/import support:
- Migration 050: import_metadata JSONB column on trees
- GET /trees/{id}/export?format=json|xml endpoint
- POST /trees/import endpoint (creates draft, resolves categories/tags)
- FlowExportEnvelope, FlowImportRequest/Response schemas
- import_metadata field on TreeResponse
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add flow export/import frontend + backend tests
Frontend:
- ExportFlowModal with JSON/XML format selection + download
- ImportFlowModal with drag-drop file picker + preview step
- rfflowParser for client-side JSON/XML .rfflow parsing
- Export buttons on editor toolbar and library action menus
- Import button on library page next to Create New
- Provenance display for imported flows in editor
- flowTransfer API client + types
Backend:
- Fix regex->pattern deprecation in export endpoint
- 12 integration tests covering export, import, round-trip,
access control, tag/category creation, version validation
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: remove XML export, JSON-only for .rfflow files
- Remove XML builder, format query param, and XML tests
- Simplify ExportFlowModal (no format picker)
- Simplify rfflowParser (JSON-only)
- Remove format field from schemas and types
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: match Flow Assist chat input to AI Assistant styling + strengthen one-question prompt
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add procedural flow support to AI chat builder (Flow Assist)
- Add procedural-specific system prompts (schema, interview protocol, response format)
- Dispatch prompts by flow_type: procedural/maintenance use flat steps schema, troubleshooting uses decision tree schema
- Parse [STEPS_UPDATE] and [INTAKE_FORM] markers in AI responses
- Add validate_generated_procedural_steps() validator
- Handle intake form extraction in AI chat import endpoint
- Add StaticStepsPreview component for procedural flow preview
- Update store and page to render correct preview by flow type
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add flow type selection to Flow Assist entry points
- CreateFlowDropdown now shows "Build with Flow Assist" under each flow type
- Library page "Flow Assist" button respects current type filter
- Clean up unused AIFlowBuilderModal references
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: update CLAUDE.md with AI chat builder and intake form learnings
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: refine assistant chat prompt for concise answers and focused questions
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: switch AI provider to Claude Sonnet 4.6 + add shift+enter hint to chat inputs
- Default AI_PROVIDER changed from gemini to anthropic
- AI_MODEL and AI_MODEL_ANTHROPIC updated to claude-sonnet-4-6
- Added "Shift + Enter for a new line" hint below all chat textareas
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: update CLAUDE.md with AI provider and chat input learnings
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: add editor-embedded Flow Assist design document
Design for replacing the standalone /ai/chat page with context-aware
AI side panels embedded in each editor (Troubleshooting + Procedural).
Covers ghost node suggestion system, output-based thresholds,
config-driven model routing, knowledge integration, and per-flow
chat persistence.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: add editor-embedded Flow Assist implementation plan
25-task plan across 9 phases covering backend foundation, frontend
infrastructure, tree/procedural editor integration, AI-assisted create,
old code removal, action-type dispatch, suggestion audit trail, and
build verification.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: use actual root node ID in orphan validation check
AI-generated trees use descriptive IDs (e.g., "verify-account-exists")
instead of "root", causing the root node to be falsely flagged as orphaned.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add config-driven AI model tier routing
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: extend AI chat session with tree_id and archived_at
Add tree_id FK (CASCADE) for editor-embedded sessions and archived_at
timestamp column to ai_chat_sessions table.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add AI suggestion audit trail table
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add action_type and focal_node_id to AI chat message API
- Add VALID_ACTION_TYPES literal and action_type/focal_node_id fields to
AIChatMessageRequest schema
- Add tree_id field to AIChatStartRequest schema for editor-embedded sessions
- Update send_message() signature with action_type and focal_node_id params
- Update start_chat_session() signature with tree_id param
- Pass new params through endpoints to service functions
- All new params have defaults so existing behavior is unchanged
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: route AI model selection through action-type config
Update get_ai_provider() to accept an optional model override parameter
(applied only to AnthropicProvider; Gemini always uses its own model).
Thread action_type-based model resolution through send_message() and
generate_final_tree() in the AI chat service.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add TypeScript types for editor-embedded AI
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add shared ContextMenu component
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add useEditorAI hook and editorAI API client
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add EditorAIPanel component with Chat and Suggestions tabs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: integrate AI panel, context menu, and ghost nodes in tree editor
- Add AI Assist panel toggle button to tree editor toolbar
- Wire EditorAIPanel alongside TreeEditorLayout with single-panel rule
- Thread onNodeContextMenu through TreeEditorLayout → FlowCanvas → FlowCanvasNode
- Add right-click context menu with Generate Branch, Explain Node, Delete actions
- Add ghost node detection (_suggestion flag) with dashed border + opacity styling
- Add Accept/Dismiss overlay buttons on ghost nodes for future suggestion handling
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: integrate AI panel, context menu, and ghost steps in procedural editor
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add AI prompt dialog and wire into CreateFlowDropdown
Replace navigation to /ai/chat with an inline AIPromptDialog modal
that collects a single prompt, generates a flow via the editor AI API,
imports it, and navigates to the editor with the AI panel open.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: add glassmorphism to AI prompt dialog + maintenance Flow Assist button
- Use .glass-card-static on AIPromptDialog card for consistent design system
- Add "Build with Flow Assist" button to maintenance section in CreateFlowDropdown
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: remove standalone Flow Assist page and old AI chat components
Remove the old /ai/chat page, AI wizard modal, and all associated
components/stores/types now replaced by the editor-embedded AI panel.
Deleted:
- AIChatBuilderPage, ai-chat/ components, aiChatStore, aiChat API, ai-chat types
- AIFlowBuilderModal, ai-builder/ components, aiFlowBuilderStore
Cleaned up:
- Router (removed /ai/chat route)
- Sidebar (removed Flow Assist nav item)
- MyTreesPage (removed AI builder modal and button)
- TreeLibraryPage (removed Flow Assist button)
- API and type barrel exports
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add delta response parsing and action-type prompt dispatch
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add AI suggestion audit trail endpoints
Create/list/resolve endpoints for tracking AI-applied changes to flows.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add APScheduler task to auto-archive stale AI chat sessions
Archives AI chat sessions with no activity for 30 days, runs daily at 3 AM.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: update project status for editor-embedded Flow Assist
- Add Editor-Embedded Flow Assist to CURRENT-STATE.md in-progress items
- Update CLAUDE.md: fix stale lessons (#41, #46), add new patterns (#47 editor AI architecture, #48 orphan validation)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: use correct model alias in AI_MODEL_TIERS standard tier
The dated model ID `claude-sonnet-4-6-20250514` was causing 502 errors.
Use the alias `claude-sonnet-4-6` which matches AI_MODEL_ANTHROPIC.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: send live flow context to AI Assist for full editor awareness
The AI panel now sends the current tree structure (troubleshooting) or
steps + intake form (procedural/maintenance) with each message. This
gives the AI full visibility into node details, questions, descriptions,
options, and intake form fields — not just the node ID.
- Backend: add flow_context param to schema, endpoint, and service
- Frontend: add getFlowContext callback to useEditorAI hook
- TreeEditorPage: passes treeStructure as flow context
- ProceduralEditorPage: passes steps + intakeForm as flow context
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: include flow name and description in AI Assist context
Both editors now send name and description alongside the flow structure,
so the AI can reference what the flow is about when responding.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: increase AI timeout to 120s and limit retries to 1
The 45s timeout was too short for generation tasks with full flow
context in the system prompt. The Anthropic SDK's default 2 retries
caused requests to hang for ~136s before failing. Now: 120s timeout
with max 1 retry = faster failure if it does timeout.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: wire AI-generated flow structures into editor stores
The useEditorAI hook was ignoring result.working_tree from AI responses,
so generated steps/trees never appeared in the editor. Now:
- useEditorAI calls onFlowUpdate when working_tree is present in response
- ProceduralEditorPage handles steps + intake form updates via replaceSteps
- TreeEditorPage handles tree structure updates via replaceTreeStructure
- Both stores have new bulk-replace methods for AI integration
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: add lessons learned for full-stack integration, Anthropic retries, model tiers
#49 Always verify frontend consumes backend response fields
#50 Anthropic SDK max_retries=1 to avoid 3× timeout
#51 AI model tier routing via settings.get_model_for_action()
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: move AI Assist panel to full-height side layout in both editors
The AI panel was nested inside the content area, only spanning the
step list / canvas section. Now it sits at the outermost flex level,
spanning the full page height alongside all content (toolbar,
collapsible sections, steps/canvas). This prevents the panel from
overlapping content and lets the editor area properly shrink.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: AI Assist panel as fixed right drawer (matching Copilot/Scratchpad)
Convert EditorAIPanel from in-flow flex child to fixed right-side drawer
overlay, same pattern as CopilotPanel and ScratchpadSidebar. The panel
is fixed at right:0 spanning full viewport height, and editor pages add
pr-[380px] padding when open so content shifts left without overlap.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: AI Assist panel sits below topbar with slide-in animation
- Panel now uses top:56px to sit below the app shell topbar instead of
covering it (matches the main-content grid cell area)
- Added slideInRight CSS animation for smooth drawer entrance
- Editor pages use dynamic paddingRight via PANEL_WIDTH constant
- ChatTab upgraded: markdown rendering, CopilotPanel-style message
bubbles, auto-focus input, Shift+Enter hint
- All borders use --glass-border for consistent glassmorphism
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: AI Assist panel as in-flow flex sibling (not fixed/overlay)
Replace fixed positioning with in-flow flex layout. The outermost div
is now a horizontal flex row: content column (flex-1 min-w-0) + panel
(w-[380px] shrink-0). When the panel opens, the content column
automatically shrinks — no padding hacks or z-index stacking needed.
This guarantees the content shifts left and stays fully visible.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: AI Copilot panel as in-flow flex sibling in session navigation pages
Changed CopilotPanel from fixed overlay to flex layout sibling so it
pushes main content instead of covering it during active sessions.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: remove duplicate CLAUDE.md lessons #47-48
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
|
||
|
chihlasm
|
babfd0c6d9 |
feat: add generate_text method to AIProvider for non-JSON responses
The AI Chat Builder needs conversational text responses, not JSON-only. Gemini's generate_json forces response_mime_type='application/json' which is incompatible. The new generate_text method omits this constraint. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |
||
|
chihlasm
|
6fc76187c0 |
fix: add diagnostic logging and increase scaffold max_tokens to 2048
The "Unterminated string" JSON parse error is likely caused by Gemini output truncation at 1024 tokens. Increases scaffold max_tokens to 2048 and adds logging for: raw response text, finish_reason (truncation detection), and JSON parse failures. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |
||
|
chihlasm
|
957f13b993 |
fix: use correct google-genai async API and remove debug endpoint
The google-genai SDK uses `client.aio.models.generate_content()` for async calls, not `client.models.generate_content_async()` which doesn't exist. Also removes the temporary /ai/provider-debug endpoint. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |
||
|
chihlasm
|
55be033ecb |
feat: add AI provider abstraction with Gemini and Anthropic support
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> |
