Tools & Integrations
Tools extend your agent's capabilities beyond conversation. They let agents call external services, run custom logic, and connect to the systems your business already runs on — all during a live conversation.
Tool Types
API Tool
Make an HTTP request to any external service — CRM, booking system, internal tool. Configure the method, URL, headers, and body, then map the response back into variables the agent can use. Great for one-off integrations where you already have an API.
Function Tool
Run a small snippet of custom logic — data transformation, formatting, calculations, or custom business rules — and return the result to the agent. Use this when you need a little glue code that doesn't justify a full API.
Integration Tool
Plug into popular platforms through a pre-built connector — CRMs, ticketing systems, payment processors, and more. Select the provider and action, supply credentials, and the integration handles the rest.
MCP Tool
Connect to any Model Context Protocol server to give your agent access to an entire catalog of tools and data sources through a single standardized connection.
Creating Tools
Tools are defined at the project level and can be shared by any agent step or playbook in the project. Each tool has a name, a description, and its type-specific configuration.
The description is the most important field. It is what the agent reads when deciding whether a tool is the right one for the situation, so write it the same way you would brief a colleague: be clear about what the tool does and when to reach for it.
Assigning Tools to Agents
After creating a tool, attach it to the agent steps or playbooks that should have access. Only attached tools are available to that agent, which lets you scope each agent to the exact capabilities it needs — nothing more, nothing less.
Built-in Tools (Conversation-flow agents)
Conversation-flow agent steps get a set of built-in capabilities you can turn on per step with a single toggle — no configuration required:
| Tool | Description |
|---|---|
| Knowledge Base | Search the project's knowledge base for relevant information |
| Buttons | Display interactive buttons in the conversation |
| Cards | Show rich cards with titles, descriptions, and images |
| Carousels | Display horizontally scrollable card sets |
| Web Search | Search the web for real-time information |
| End Conversation | Gracefully end the current conversation when the goal is met |
Built-in Tools (Agentic mode)
Agentic-mode projects ship with a different set of built-in tools — the platform primitives the harness exposes to the model on every turn. Some are always available; the UI cards are opt-in via behaviourConfig.agentHarness.uiBuiltins.
Always available
| Tool | Purpose | Required args |
|---|---|---|
| set_state | Merge structured fields into session state. Persists for the next turn and appears in the snapshot. | variables (object) |
| kb_lookup | Query a structured KB table by name + equality filters. Returns matching rows. | table |
| kb_search | Semantic search across the project KB (documents, PDFs, URLs). Returns top-k chunks with citations. | query |
| show_buttons | Show clickable buttons. The user can tap one (sends label as next text turn) or type freely. | buttons[{label}] |
| request_human | Flag the conversation for human takeover. Sets needs_human=true and triggers the notify hook. | — (reason optional) |
| handoff | Hand off to a registered playbook (specialist sub-agent). Bounded by allowedPlaybookIds + maxHandoffs. | targetPlaybookId |
UI builtins (opt-in)
Each defaults to false. Enable per project via behaviourConfig.agentHarness.uiBuiltins.<name>: true. Disabled tools never appear in the model's tool catalog (zero token cost).
| Tool / Toggle | Purpose | Required args |
|---|---|---|
| show_badge uiBuiltins.showBadge | Single-line status badge. Use for severity, classification, status pill. | label |
| show_tip uiBuiltins.showTip | Guidance card with recommended actions and things to avoid. Safety advice, troubleshooting hints. | title |
| show_summary uiBuiltins.showSummary | Labeled-fields summary card. Ticket recap, order confirmation, profile overview. | title, fields[] |
| show_action_steps uiBuiltins.showActionSteps | Ordered checklist. Each step has title + optional description + status. | steps[] |
| show_contact_card uiBuiltins.showContactCard | Person or team contact card with phone/email/hours and arbitrary alternate channels. | name |
Phase-level allowedToolIds still applies on top — a tool must be enabled at the project AND allowed by the active phase to surface to the model on a given turn. See Agentic Mode.
Choosing a Model
Every agent, playbook, and prompt step lets you pick the model that powers it, along with temperature, max tokens, and retry behavior. Conduet supports a curated selection of the latest frontier models so you can balance speed, cost, and reasoning ability for each step. Start with the recommended defaults and tune from there as you see how your agent performs.