README
¶
Azure OpenAI — Tool Calling
Multi-round tool calling against an Azure OpenAI deployment. The model is given two tools and Chronos drives the full tool-call loop until a final answer is produced.
What it demonstrates
- Wiring
tool.Registrydefinitions into a provider-agnosticmodel.ChatRequest.Toolsslice - The tool-call loop: detect
StopReason == model.StopReasonToolCall, execute each requested tool, appendRoleToolresult messages, and re-ask — bounded bymaxToolRounds - Two tools: a
calculator(add/sub/mul/div) and alookup(canned facts)
The loop is identical to the one in examples/graph_with_llm; only the
provider construction is Azure-specific.
Run
export AZURE_OPENAI_API_KEY=<your-azure-api-key>
export AZURE_OPENAI_ENDPOINT=https://<your-resource>.openai.azure.com
export AZURE_OPENAI_DEPLOYMENT=<your-deployment-name> # must support function calling
export AZURE_OPENAI_API_VERSION=2024-10-21
go run ./examples/azure_tools/main.go
Without AZURE_OPENAI_API_KEY the example prints the required variables and
exits 0 — no network call.
Tests
main_test.go is fully offline: it unit-tests the pure calculate and
lookup helpers and the tool.Registry wiring. It never contacts Azure.
go test ./examples/azure_tools/...
See also: examples/azure (chat/streaming) and
examples/azure_rag (RAG).
Documentation
¶
Overview ¶
Azure OpenAI tool-calling example for Chronos.
This example shows how to drive multi-round tool calling against an Azure OpenAI deployment. The model is given two tools — a calculator and a small knowledge lookup — and the example runs the full tool-call loop: it sends the request with tool definitions, and whenever the model responds with StopReasonToolCall it executes the requested tools, feeds the results back, and asks again, repeating until the model produces a final answer.
The same loop pattern works for any provider (see examples/graph_with_llm), because tools are wired through the provider-agnostic model.ChatRequest.
Prerequisites:
- An Azure OpenAI resource with a chat model deployment (e.g. gpt-4o-mini) that supports function calling
- Go 1.24+
Set the following environment variables before running:
export AZURE_OPENAI_API_KEY=<your-azure-api-key> export AZURE_OPENAI_ENDPOINT=https://<your-resource>.openai.azure.com export AZURE_OPENAI_DEPLOYMENT=<your-deployment-name> export AZURE_OPENAI_API_VERSION=2024-10-21
Run:
go run ./examples/azure_tools/main.go
If AZURE_OPENAI_API_KEY is not set the example prints the required variables and exits cleanly (0) without making any network call.
Source Files