Appearance
dgi.thought-processing-synthesize
Synthesize results — calls the org's LLM via ai.chat to produce the final answer
Synthesize results - merge reasoning and skill outputs
Overview
| Property | Value |
|---|---|
| Workflow type | Linear |
| Library | App-dgi |
| Version | 1.0 |
Input Schema
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
reasoning_output | json | No | — | Reasoning output |
skill_results | json | No | — | Skill execution results |
dag_structures | json | No | — | DAG structures from skills |
message | string | No | — | Original user message |
history | json | No | — | Conversation history slice [{role, content}, …] |
organization_uuid | uuid | No | — | Org UUID — propagated so ai.chat can resolve the org's default AIProviderConfig when no connection/provider uuid is supplied |
connection_uuid | uuid | No | — | Optional explicit AI CloudConnection UUID; highest precedence in ai.chat resolution |
ai_provider_config_uuid | uuid | No | — | Optional AIProviderConfig UUID; second precedence (after connection_uuid). When neither is supplied, ai.chat resolves the org's default AIProviderConfig |
system_prompt | string | No | — | Optional system instruction override |
model | string | No | — | Optional model override |
max_tokens | integer | No | — | Optional max tokens (default 1024) |
temperature | json | No | — | Optional temperature (default 0.7) |
working_set | json | No | — | Conversation Working Set carried from the prior turn (ADR 0001): slots, candidates, in-flight composition |
confirmed_action | json | No | — | Resolution of a prior pending_action (drives confirm=true in the reasoning loop) |
provided_inputs | json | No | — | Values the user filled into a prior pending_input form |
chat_conversation_uuid | uuid | No | — | conversation_uuid of the DGI chat session that originated this turn (session-timeline provenance) |
intent_analysis | json | No | — | Fast/deep gate verdict from analyze_intent: |
max_reasoning_turns | integer | No | — | Optional tool-loop turn budget for this turn; clamped to [6, 48], default 24 |
proactive | boolean | No | — | Proactive mode: ask the model to suggest 2-3 next actions after answering |
workflow_run_id | string | No | — | Engine-stamped DAG step run id |
Output Schema
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
reasoning_output | json | No | — | — |
skill_results | json | No | — | — |
dag_structures | json | No | — | — |
message | string | No | — | — |
history | json | No | — | — |
organization_uuid | uuid | No | — | — |
connection_uuid | uuid | No | — | — |
ai_provider_config_uuid | uuid | No | — | — |
system_prompt | string | No | — | — |
model | string | No | — | — |
max_tokens | integer | No | — | — |
temperature | json | No | — | — |
working_set | json | No | — | — |
confirmed_action | json | No | — | — |
provided_inputs | json | No | — | — |
intent_analysis | json | No | — | — |
max_reasoning_turns | integer | No | — | — |
proactive | boolean | No | — | — |
chat_conversation_uuid | uuid | No | — | Originating DGI chat session (session-timeline provenance); carried via **context.state_data |
workflow_run_id | string | No | — | — |
final_answer | string | No | — | The user-facing answer |
confidence | json | No | — | Real confidence {overall: float} computed from the turn's outcome |
llm_called | boolean | No | — | — |
model_used | string | No | — | — |
provider | string | No | — | — |
input_tokens | integer | No | — | — |
output_tokens | integer | No | — | — |
actions_taken | json | No | — | — |
reasoning_trace | json | No | — | — |
suggestions | json | No | — | Proactive next-action suggestions [{label, prompt}] emitted via suggest_next_actions |
pending_action | json | No | — | — |
pending_input | json | No | — | — |
renderer | string | No | — | — |
render | json | No | — | — |
status | string | No | — | complete |
incomplete_reason | string | No | — | When incomplete: turns_exhausted |
reasoning_child_workflow_ref | string | No | — | Async child workflow that ran the long reasoning loop |
failure_reason | string | No | — | — |
failure_type | string | No | — | — |
failed_action | string | No | — | — |
failed_at_state | string | No | — | — |
failed_step | string | No | — | — |
failed_layer | string | No | — | — |
error | string | No | — | — |
error_type | string | No | — | — |
reasoning_state | json | No | — | — |
reasoning_deps | json | No | — | — |
reasoning_seq | integer | No | — | — |
reasoning_done | boolean | No | — | — |
reasoning_result | json | No | — | — |
States
| State | Initial | Terminal | Success | Auto-advance | Description |
|---|---|---|---|---|---|
pending | Yes | No | — | execute | — |
waiting_reasoning | No | No | — | — | — |
completed | No | Yes | Yes | — | — |
failed | No | Yes | No | — | — |
State Diagram
Transitions
| From | Action | To | Description |
|---|---|---|---|
pending | execute | waiting_reasoning | — |
waiting_reasoning | child_done | completed | — |
waiting_reasoning | child_failed | failed | — |
* (any state) | fail | failed | — |
API Usage
bash
POST /api/workflows/start
Content-Type: application/json
{
"workflow_type": "dgi.thought-processing-synthesize",
"initial_data": {
"reasoning_output": "value",
"skill_results": "value",
"dag_structures": "value",
"message": "value"
}
}