Loquent · AI Operations

AI Operations Inventory

Every place the codebase calls an AI model — what it does, what goes in and out, which call shape it uses, and the source file. Use it as a map for adding new call sites and as a pre-flight checklist for migrating off aisdk onto rig.

Total call sites

OpenAI REST

Transcription

single-shot / text

single-shot / json

agentic

streaming + agentic

Assistant

Current stack

What each concern uses today, and what the rig migration has to replace.

Concern	Today	Migration target
LLM SDK	`aisdk::core::LanguageModelRequest` via `Openrouter::<DynamicModel>`	Replace builder + `generate_text()`/`stream_text()` with `rig` agent calls
Model resolution	`crate::mods::ai::resolve_model(AiArea::*)` → model id string	Keep `AiArea` enum, swap resolver target
Structured output	`schemars::JsonSchema` on a private struct + `generate_text()` returning typed JSON	Map to `rig` structured-output / extractor APIs
Tools	`.with_tool(t)` + `step_count_is(N)` stop condition	Map to `rig` tool registration + agent loop
Streaming	`.stream_text()` + `LanguageModelStreamChunkType`	Map to `rig` streaming agent
Usage logging	`/add-ai-logging` skill — required at every call site	Preserve identical fields/args
Transcription	OpenAI REST via `reqwest::multipart` (not `aisdk`)	Out of scope — leave as-is

Mode legend

Each call site is tagged with one of these. Look for the colored dot.

OpenAI REST

Raw multipart upload via reqwest. Used only for transcription.

single-shot / text

One prompt, plain-text answer. No tools, no schema.

single-shot / json

One prompt, typed JSON answer via schemars::JsonSchema.

agentic

Tool-using loop with a step_count_is(N) stop.

streaming + agentic

Chunked output with tool calls. Only the in-app assistant.

wrapper

Orchestration around another AI op — not a model call itself.

User-facing pipelines

Where each call site sits in the larger flow. Each box is one AI op or a small group of related ops.

Post-call pipeline

REST

Transcribe

audio → transcript

→

json

Identify speakers

label → role

→

textjson

Analyze · Summarize · Auto-tag

analyzers + summary + tags

→

json

Enrich · Tasks · Plans

contact fields, tasks, plans

Inbound SMS pipeline

agentic

Text agent reply

single reply or 3 suggestions

→

json

Memory + tasks + enrichment

batched per contact

→

json

Create plans from SMS

match templates → instantiate

Plan execution loop

json

Instantiate plan

title + description from template

→

agentic

Execute plan (loop)

custom plan tools · step_count_is(25) · fallback model

In-app assistant

streaming + agentic

Run assistant

full tool suite · MAX_TOOL_STEPS · streamed chunks

→

text

Generate title

first messages → title

Transcription

Raw OpenAI REST. Outside the aisdk → rig migration scope.

Transcribe audio

OpenAI REST

Default OpenAI transcription wrapper.

I/O: audio bytes → TranscriptionResponse

src/mods/openai/utils/transcribe_audio_util.rs

Transcribe audio (options)

OpenAI REST

Custom transcription request params.

I/O: bytes + TranscriptionRequest → TranscriptionResponse

src/mods/openai/utils/transcribe_audio_with_options_util.rs

Transcribe audio (diarized)

OpenAI REST

Speaker-labeled via gpt-4o-transcribe-diarize.

I/O: bytes → DiarizedTranscriptionResponse

src/mods/openai/utils/transcribe_audio_diarized_util.rs

Identify speakers

single-shot / json

Map diarized labels to AI vs. caller.

I/O: diarized transcript → HashMap<label, role>

AiArea: IdentifySpeakers

src/mods/openai/utils/identify_speakers_util.rs

Process Twilio recording

wrapper

Orchestrate download + transcription.

I/O: TwilioRecordingRequest → transcription

src/mods/twilio/utils/process_twilio_recording_util.rs

Call Analysis

Post-call pipeline. Runs after a recording is transcribed.

Analyze call text

single-shot / text

Run an analyzer's custom prompt on a transcription.

I/O: prompt + transcription → free text

AiArea: AnalyzeCall

src/mods/analyzer/utils/analyze_call_text_util.rs

Auto-tag contact

single-shot / json

Pick org tags relevant to a call.

I/O: summary + tag list → tag_ids[]

AiArea: AutoTagContact

src/mods/contact/services/auto_tag_contact_from_call_service.rs

Enrich contact (call)

single-shot / json

Fill stub contact fields from transcription.

I/O: transcription → {first_name, last_name, company, …}

AiArea: EnrichContact

src/mods/contact/services/enrich_contact_service.rs

Summarize call

single-shot / json

Markdown summary for an interaction note.

I/O: transcription → sectioned markdown

AiArea: SummarizeCall

src/mods/contact/services/notes/summarize_call_for_note_service.rs

Extract tasks (call)

single-shot / json

Extract actionable tasks from a call.

I/O: transcription → tasks[{title,…}]

AiArea: ExtractTasks

src/mods/task/services/create_tasks_from_call_service.rs

Create plans (call)

single-shot / json ×N

Match post_call-trigger templates and spawn plans.

I/O: call + templates → relevance + instantiated plans

AiArea: AssessPlanTemplateStartCondition InstantiatePlan

src/mods/plan/services/create_plans_from_call_service.rs

Contact memory & enrichment

Message-pipeline counterparts to the call-pipeline enrichment ops.

Update contact memory

single-shot / json

Refresh memory note from new call/SMS context.

I/O: profile + transcript/messages → updated memory note

AiArea: UpdateSystemNote UpdateContactMemoryFromMessages

src/mods/contact/services/notes/update_contact_memory_service.rs

Analyze contact messages

single-shot / json

Extract memory + tasks from unprocessed SMS batch.

I/O: message batch → {memory_updates, tasks}

AiArea: ExtractTasksFromMessages

src/mods/contact/services/ai/analyze_contact_messages_service.rs

Enrich contact (messages)

single-shot / json

Fill contact fields from SMS history.

I/O: message history → contact attributes

AiArea: EnrichContactFromMessages

src/mods/contact/services/ai/enrich_contact_from_messages_service.rs

Plans

Template matching, instantiation, and the agentic execution loop.

Create plans from SMS

single-shot / json ×N

Match new_sms-trigger templates and instantiate.

I/O: sms + templates → instantiated plans

AiArea: AssessPlanTemplateStartCondition InstantiatePlan

src/mods/plan/services/create_plans_from_sms_service.rs

Create plan from template

single-shot / json

Generate contextual title/description for a contact.

I/O: contact ctx + template → {title, description}

AiArea: InstantiatePlan

src/mods/plan/services/create_plan_from_template_service.rs

Execute plan

agentic

Agentic loop driving plan actions — custom plan tools, step_count_is(25), fallback model.

I/O: plan state → tool calls + log entries

AiArea: ExecutePlan ExecutePlanFallback

src/mods/plan/services/execute_plan_service.rs

Text Agent

Inbound SMS handlers — direct reply or suggestions for review.

Generate single reply

agentic

One direct reply for an inbound message — query_knowledge, step_count_is(5).

I/O: conversation + agent config → text reply

AiArea: TextAgentSingleReply

src/mods/text_agent/services/generate_text_agent_single_reply_service.rs

Generate suggestions

agentic

Three reply candidates for human review.

I/O: conversation + agent config → TextAgentSuggestions{items[]}

AiArea: TextAgentSuggestions

src/mods/text_agent/services/generate_text_agent_suggestions_service.rs

Assistant

In-app agentic chat. The only streaming call site.

Run assistant

streaming + agentic

In-app agentic chat — full assistant tool suite, MAX_TOOL_STEPS.

I/O: messages + page ctx → streamed chunks + tool calls

AiArea: Assistant

src/mods/assistant/services/assistant_service.rs

Generate conversation title

single-shot / text

Persist a short title for an assistant chat.

I/O: first messages → title string

AiArea: AssistantTitle

src/mods/assistant/services/generate_title_service.rs

Knowledge

KB query — used as a tool inside the text-agent and assistant loops.

Query knowledge

single-shot / text

Extract concise answer from KB documents.

I/O: query + KB docs → answer text

AiArea: KnowledgeQuery

src/mods/agent/services/handle_query_knowledge_tool_call_service.rs

Dashboard & Reports

Briefing cards and the daily org report.

Dashboard briefing

single-shot / json

Daily structured briefing card.

I/O: user data ctx → JSON briefing

AiArea: DashboardBriefing

src/mods/dashboard/services/generate_dashboard_briefing_service.rs

Detailed briefing

agentic

Investigative briefing with tool calls — dashboard tools, step_count_is(10).

I/O: scope → JSON briefing

AiArea: DashboardDetailedBriefing

src/mods/dashboard/services/generate_detailed_briefing_service.rs

Daily report

single-shot / text

End-of-day org report.

I/O: calls + messages window → report text

AiArea: GenerateReport

src/mods/report/services/generate_daily_report_service.rs

AI Builder (authoring UX)

All consume Session and run via Dioxus #[post] server functions. All return structured JSON via JsonSchema.

Generate instructions

single-shot / json

Instructions from a description.

I/O: AiBuilderGenerateRequest → AiBuilderResponse

AiArea: GenerateInstructions

src/shared/ai_builder/api/generate_api.rs

Edit instructions (preset)

single-shot / json

Apply preset action to instructions.

I/O: AiBuilderEditRequest → AiBuilderResponse

AiArea: EditInstructions

src/shared/ai_builder/api/edit_api.rs

Edit instructions (custom)

single-shot / json

Free-text edit of instructions.

I/O: AiBuilderCustomEditRequest → AiBuilderResponse

AiArea: CustomEditInstructions

src/shared/ai_builder/api/custom_edit_api.rs

Generate text agent

single-shot / json

All four text-agent fields.

I/O: TextAgentGenerateRequest → TextAgentAiResponse

AiArea: GenerateInstructions

src/shared/ai_builder/api/text_agent_generate_api.rs

Edit text agent (preset)

single-shot / json

Preset action across four fields.

I/O: TextAgentEditRequest → TextAgentAiResponse

AiArea: EditInstructions

src/shared/ai_builder/api/text_agent_edit_api.rs

Edit text agent (custom)

single-shot / json

Free-text edit across four fields.

I/O: TextAgentCustomEditRequest → TextAgentAiResponse

AiArea: CustomEditInstructions

src/shared/ai_builder/api/text_agent_custom_edit_api.rs

Generate plan template

single-shot / json

Plan template description + actions.

I/O: PlanTemplateGenerateRequest → PlanTemplateAiResponse

AiArea: GenerateInstructions

src/shared/ai_builder/api/plan_template_generate_api.rs

Migration plan

Suggested order for the aisdk → rig swap. Smallest blast radius first.

1

single-shot / text 4 sites

Simplest shape. Start with Analyze call text, Query knowledge, Generate conversation title, Daily report.
2

single-shot / json 18 sites

Schema bridge — most of the inventory. Validates that JsonSchema → rig extractor mapping holds across shapes.
3

agentic (non-streaming) 4 sites

Execute plan, Detailed briefing, Text agent single reply, Text agent suggestions. Confirms tool-loop semantics + step-count parity.
4

streaming 1 site

Largest blast radius — the in-app assistant. Save for last.

Notes

src/mods/assistant/tools/** are tool definitions registered into the assistant loop — not separate AI call sites. Same for src/mods/ai/tools/** and the per-domain tool builders consumed by Execute plan and Detailed briefing.
Wrappers with no direct AI call: src/mods/widget/routes/widget_ws_route.rs (→ Generate single reply), src/mods/text_agent/api/suggestions_api.rs (→ Generate suggestions), src/mods/text_agent/services/handle_inbound_message_for_text_agent_service.rs (→ suggestions service).
Every new call site must add usage logging — see the /add-ai-logging skill.