Chat and agent modes

Overview

Ottili AI — the AI control surface of Ottili ONE — can fundamentally be used in two ways:

• Chat mode*: You hold a conversation with the AI. You ask a question or give a task, and the AI replies in the same context. This is the mode of the assistant inside Ottili ONE and of the OpenAI-compatible Chat Completions* endpoint (/v1/chat/completions).
• Agent mode*: You hand Ottili AI a goal. The AI breaks it into work, assigns DWE specialists* (role-based agents), calls tools* when needed, and only runs critical steps after your approval*. The building blocks of this mode — missions, DWE, tool use, approvals — are live; a dedicated, packaged "Agent Mode" surface is in development.

Both modes run on the same Ottili models (Ottili Vale*, Ottili Cairn*) and the adaptive router Ottili Auto*, are billed through the same credit system, and execute inside the same company context.

Chat mode (live)

Chat mode is generally available (General Availability, Live*). It is the most direct way to use Ottili AI:

• In product*: Via the AI icon at the bottom-right or the AI panel inside a record (customer, order, quote).
• Via the API*: Via the OpenAI-compatible Chat Completions endpoint POST /v1/chat/completions with model, messages and optional stream. The programmatic surface is part of the public Ottili AI API (Public Beta).

Typical chat-mode uses:

• Summarise (customer record, order history, thread)
• Draft (email replies, quotes, messages, documents)
• Plan (break a complex task into steps)
• Analyse (insights from business data)
• Act (trigger Flows and module actions — always behind an approval)

In chat mode you control the pace and direction: you send a message, review the response, correct if needed. Any action that touches the outside world (send a message, create an order, publish content) goes through the approval queue first.

Agent mode (building blocks live, surface in development)

In agent mode you hand over not a single prompt but a goal* — a mission*. The Dynamic Workforce Engine (DWE)* splits the mission into work and assigns DWE specialists* that carry it forward. This follows the same AI capability flow as the public Ottili AI page:

1. Model*: Vale or Cairn execute the individual work.

2. Routing*: Ottili Auto picks the right model per step.

3. Mission*: The handed-over goal structures the work.

4. DWE specialists*: Role-based agents carry the work forward.

5. Tools*: Specialists call tools across the Unified API business.* and ld3.* namespaces.

6. Approvals*: Critical, externally-visible actions are approval-first.

7. Credits & cost control*: Every request is reserved and deducted.

8. Audit*: Every action is resolved to one company and logged.

Status of the building blocks (source: config/product_truth/ai_feature_status.yaml):

• Missions* — Live
• DWE specialists* — Live
• Tool use* — Live
• Approvals* — Live (enforced through Ottili Console and Ottili Core)
• Credits, cost control, audit* — Live (Core-backed)
• Memory* — Planned (not yet evidenced as live)

A dedicated, standalone "Agent Mode" surface (its own entry point, sessions, integrations with Business Hub, Flows, LD3) is in development / planned* and not yet generally available. The agentic way of working itself builds on the already-live capabilities listed above.

Which mode when

• Chat mode* when: you want fast, conversational help — summarising, drafting, ad-hoc questions, single steps.
• Agent mode* when: a task breaks into multiple steps, specialist roles make sense, or you want the AI to carry work forward over time — always approval-controlled.

In both modes: Ottili AI acts only inside your company context. Nothing leaves your company for an external effect without a required approval being granted.

Models and routing in both modes

Whichever mode you use, model and routing determine behaviour:

• Ottili Vale* — efficient, everyday, cost-effective (Live/GA).
• Ottili Cairn* — premium, deeper reasoning for serious engineering and analysis work (Live/GA).
• Ottili Auto* — the adaptive router that picks Vale or Cairn per request (Live/GA).
• Ottili Spire* — reserved flagship, not yet available.

How models, routing and availability work is described in [Ottili AI API — Models](/docs/ottili-ai-api-models).

Cost and approvals

Both modes are prepaid on the shared, company-scoped credit wallet; per request, cost is reserved and deducted. Approval rules and budgets are configured under Settings → AI. How credits and plans relate is described in [Understand plans and credits](/docs/understand-plans-and-credits).

Status at a glance

How Ottili ONE publicly marks readiness is described in [Understand feature-status labels](/docs/understand-feature-status-labels).

Related articles

• [What is Ottili ONE?](/docs/what-is-ottili-one)
• [Ottili AI](/docs/ottili-ai)
• [Ottili AI API](/docs/ottili-ai-api)
• [Ottili AI API — Models](/docs/ottili-ai-api-models)
• [Ottili AI API — Tool calls](/docs/ottili-ai-api-tool-calls)
• [Understand feature-status labels](/docs/understand-feature-status-labels)
• [Understand plans and credits](/docs/understand-plans-and-credits)