# Quote from a price list ## Requirement "Give a quote from our price list using our discount formula." ## Pieces - **Custom data**: a collection `prices`, one record per item. - **Custom tool**: `quote`, the discount formula, handed the price record and the order and answering the total. - **Flow**: query the price, then quote. - **The flow as a tool**, attached to a node, on a published agent. ## The shape to avoid The prices and the formula in the prompt. The model does arithmetic badly and inconsistently, the prices go stale, and nobody can see which price a quote was made from. Here the price is a record, the formula is code that runs the same way every time, and the flow ties the two together. ## Steps Ids from an earlier step are written as `{{step_N.field}}`; the test that runs this page substitutes them. 1. The price list. `create_collection` ```json { "name": "Prices", "slug": "prices", "fields": [ {"key": "sku", "label": "SKU", "type": "text", "required": true}, {"key": "item", "label": "Item", "type": "text", "required": true}, {"key": "unit_price", "label": "Unit price", "type": "number", "required": true} ] } ``` 2. A price. `create_record` (`collection_id` = `{{step_1.id}}`) ```json {"values": {"sku": "CHAIR-01", "item": "Office chair", "unit_price": 120}} ``` 3. Another. `create_record` (`collection_id` = `{{step_1.id}}`) ```json {"values": {"sku": "DESK-01", "item": "Standing desk", "unit_price": 450}} ``` 4. The formula. It is handed the matching price records, the quantity and the customer type, and answers a value. `tool_id` in the response is what the flow step names. `create_custom_tool` ```json { "name": "quote", "description": "The total for a quantity of one item, with the business discount applied.", "purpose": "Apply the discount formula: businesses ordering ten or more get 10% off; everyone else pays the list price.", "source_code": "def transform(arguments):\n prices = arguments.get(\"prices\") or []\n if not prices:\n return {\"found\": False}\n unit = float(prices[0][\"unit_price\"])\n quantity = int(arguments.get(\"quantity\") or 1)\n business = arguments.get(\"customer_type\") == \"business\"\n discount = 0.10 if business and quantity >= 10 else 0.0\n total = unit * quantity * (1.0 - discount)\n return {\"found\": True, \"item\": prices[0][\"item\"], \"unit_price\": unit, \"quantity\": quantity, \"discount\": discount, \"total\": total}\n", "input_schema": { "type": "object", "properties": { "prices": {"type": "array", "description": "The matching records of prices."}, "quantity": {"type": "integer"}, "customer_type": {"type": "string", "enum": ["private", "business"]} }, "required": ["prices", "quantity", "customer_type"] }, "output_schema": { "type": "object", "properties": { "found": {"type": "boolean"}, "item": {"type": "string"}, "unit_price": {"type": "number"}, "quantity": {"type": "integer"}, "discount": {"type": "number"}, "total": {"type": "number"} } }, "declared_reach": [] } ``` 5. Publish it. `publish_custom_tool` (`custom_tool_id` = `{{step_4.id}}`) ```json {} ``` 6. The flow. `create_flow` ```json {"name": "quote_from_price_list"} ``` 7. Entry step: the price for the requested SKU, with the collection's query tool. The flow's inputs are in scope as `$ref:trigger.`. `add_step` (`flow_id` = `{{step_6.id}}`) ```json {"tool_id": "{{step_1.query_tool_id}}", "param_overrides": {"sku": "$ref:trigger.sku", "limit": 1}} ``` 8. Then the formula, over the records the query answered. `add_step` (`flow_id` = `{{step_6.id}}`) ```json { "tool_id": "{{step_4.tool_id}}", "param_overrides": { "prices": "$ref:{{step_7.id}}.records", "quantity": "$ref:trigger.quantity", "customer_type": "$ref:trigger.customer_type" } } ``` 9. Query, then quote. The flow's result is the last step's output: the quote. `add_flow_edge` (`flow_id` = `{{step_6.id}}`) ```json {"source_step_id": "{{step_7.id}}", "target_step_id": "{{step_8.id}}"} ``` 10. The flow as one tool. `create_tool` ```json { "name": "quote", "description": "A quote for a quantity of one item, discount applied.", "source": "builtin", "config": {"flow_id": "{{step_6.id}}"}, "input_schema": { "type": "object", "properties": { "sku": {"type": "string"}, "quantity": {"type": "integer"}, "customer_type": {"type": "string", "enum": ["private", "business"]} }, "required": ["sku", "quantity", "customer_type"] } } ``` 11. An agent publishes only with an AI profile. `create_llm_settings` ```json {"name": "fast", "provider": "openai", "model": "gpt-4o-mini"} ``` 12. The profile. `create_profile` ```json {"name": "default", "text": {"llm_settings_id": "{{step_11.id}}"}} ``` 13. The agent. `create_agent` ```json {"name": "Sales"} ``` 14. Its draft. `get_draft` (`agent_id` = `{{step_13.id}}`) ```json {} ``` 15. The node, with the quote tool. `add_node` (`graph_id` = `{{step_14.graph_id}}`) ```json { "instructions": "You quote prices. Ask which item, how many, and whether the caller is a business or a private customer, then call quote and read back the total. Never compute a price yourself.", "tool_ids": ["{{step_10.id}}"] } ``` 16. Named as the root. `set_root` (`graph_id` = `{{step_14.graph_id}}`, `node_id` = `{{step_15.id}}`) ```json {} ``` 17. The profile on the draft. `edit_draft` (`version_id` = `{{step_14.id}}`) ```json {"ai_profile_id": "{{step_12.id}}"} ``` 18. Published. `publish_version` (`version_id` = `{{step_14.id}}`) ```json {} ``` ## What the agent sees One tool, `quote`, with `sku`, `quantity` and `customer_type`, answering the custom tool's result under `result`: the item, the unit price, the discount and the total. The price list and the formula are invisible from the conversation. ## Variations - **The formula changes**: edit the custom tool's draft (`edit_custom_tool_draft`) and publish again; the flow, the tool and the agent are untouched. - **A price changes**: `update_record`; nothing is republished. - **The item is named, not coded**: query on `item` instead of `sku`, and let the node instructions ask the model to pick the SKU from `query_prices` first.