Skip to content

Switchyard lets LLM applications route traffic across models and providers while preserving native OpenAI and Anthropic API compatibility - enabling flexible model selection, benchmarking, and cost/performance optimization.

Rust

docs: add single-provider coding agent recipes (#870)

ryan-lempka1dd898a1f7e5
diff --git a/docs/index.md b/docs/index.md
index b477c1eeb..b8971e066 100644
--- a/docs/index.md
+++ b/docs/index.md
@@ -8,6 +8,7 @@ It supports OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages.
| Goal | Path | Start here |
|---|---|---|
| Run Switchyard as a standalone proxy for API clients | Server Path | [Build and run the Rust server](getting_started.md#server-path) |
+| Route Codex or Claude Code using your existing login | Server Path | [Single-provider coding agents](recipes/single_provider_coding_agents.md) |
| Add Switchyard routing to a Rust application | Library Path | [`switchyard-libsy`](../crates/libsy/README.md) |
| Add Switchyard routing to NeMo Relay | Native Plugin Path | [Use Switchyard with NeMo Relay](integrations/nemo_relay.md) |
| Point the pi coding agent at the standalone proxy | Server Path | [Use Switchyard with pi](integrations/pi.md) |
diff --git a/docs/recipes/single_provider_coding_agents.md b/docs/recipes/single_provider_coding_agents.md
new file mode 100644
index 000000000..6ede23313
--- /dev/null
+++ b/docs/recipes/single_provider_coding_agents.md
@@ -0,0 +1,165 @@
+# Interactive coding agents with single-provider auth
+
+Route Codex or Claude Code between models from one provider using your saved CLI
+login. Every target, including the classifier, must use that provider.
+
+These [composite routing](../routing_algorithms/composite_routing.md) recipes use a
+classifier to choose the starting tier for each user turn. Stage adjusts routing
+during tool calls. [`forward_auth`](../reference/toml_schema.md#llm_clientsname)
+passes your login through to the models.
+
+Choose one recipe. Each uses local port `4123`.
+
+## Codex with OpenAI
+
+Sign in with `codex login`. Terra classifies, and Stage routes between Sol and Luna.
+
+This recipe needs Codex classifier support added after v0.3.0. Install from a
+checkout of current `main`:
+
+```bash
+cargo install --locked --path crates/switchyard-server
+```
+
+Save as `codex-routing.toml`:
+
+```toml
+schema_version = 1
+
+[llm_clients.chatgpt]
+format = "openai_responses"
+base_url = "https://chatgpt.com/backend-api/codex"
+forward_auth = true
+
+[targets.capable]
+id = "gpt-5.6-sol"
+llm_client = "chatgpt"
+
+[targets.efficient]
+id = "gpt-5.6-luna"
+llm_client = "chatgpt"
+
+[targets.judge]
+id = "gpt-5.6-terra"
+llm_client = "chatgpt"
+extra_body = { store = false, stream = true }
+omit_body_fields = ["max_output_tokens"]
+
+[routes.switchyard]
+id = "switchyard"
+type = "composite"
+
+[routes.switchyard.classifier]
+target = "judge"
+base_threshold = 0.5
+classify_trigger = "user_turn"
+
+[routes.switchyard.stage]
+capable_target = "capable"
+efficient_target = "efficient"
+confidence_threshold = 0.5
+```
+
+Start the server:
+
+```bash
+switchyard-server --config codex-routing.toml --dry-run
+switchyard-server --config codex-routing.toml --host 127.0.0.1 --port 4123
+```
+
+Create `~/.codex/switchyard.config.toml` (or place it beside your `config.toml` if
+you use a custom Codex config directory):
+
+```toml
+model = "switchyard"
+model_provider = "switchyard"
+
+[model_providers.switchyard]
+name = "Switchyard"
+base_url = "http://127.0.0.1:4123/v1"
+wire_api = "responses"
+requires_openai_auth = true
+```
+
+In another terminal, launch Codex:
+
+```bash
+codex --profile switchyard
+```
+
+## Claude Code with Anthropic
+
+Sign in to Claude Code with your Claude account. Haiku classifies, and Stage routes
+between Opus and Sonnet.
+
+Install the latest release:
+
+```bash
+cargo install --locked switchyard-server
+```
+
+Save as `claude-routing.toml`:
+
+```toml
+schema_version = 1
+
+[llm_clients.anthropic]
+format = "anthropic_messages"
+base_url = "https://api.anthropic.com"
+forward_auth = true
+
+[targets.capable]
+id = "claude-opus-5-5"
+llm_client = "anthropic"
+
+[targets.efficient]
+id = "claude-sonnet-5-5"
+llm_client = "anthropic"
+
+[targets.judge]
+id = "claude-haiku-4-5-20251001"
+llm_client = "anthropic"
+
+[routes.switchyard]
+id = "switchyard"
+type = "composite"
+
+[routes.switchyard.classifier]
+target = "judge"
+base_threshold = 0.5
+classify_trigger = "user_turn"
+
+[routes.switchyard.stage]
+capable_target = "capable"
+efficient_target = "efficient"
+confidence_threshold = 0.5
+```
+
+Start the server:
+
+```bash
+switchyard-server --config claude-routing.toml --dry-run
+switchyard-server --config claude-routing.toml --host 127.0.0.1 --port 4123
+```
+
+In another terminal, launch Claude Code:
+
+```bash
+env -u ANTHROPIC_API_KEY -u ANTHROPIC_AUTH_TOKEN \
+ ANTHROPIC_BASE_URL=http://127.0.0.1:4123 \
+ ANTHROPIC_DEFAULT_OPUS_MODEL=switchyard \
+ ANTHROPIC_DEFAULT_SONNET_MODEL=switchyard \
+ ANTHROPIC_DEFAULT_HAIKU_MODEL=switchyard \
+ claude --model switchyard
+```
+
+Use your saved Claude login, without an `apiKeyHelper` or gateway token.
+
+## Check the routing
+
+Send a prompt, then check which models handled the answer and classification:
+
+```bash
+curl -s http://127.0.0.1:4123/v1/stats \
+ | jq '{answers: .models, classifier: .classifier.models}'
+```
diff --git a/mkdocs.yml b/mkdocs.yml
index 5cd2a986c..57cf32d62 100644
--- a/mkdocs.yml
+++ b/mkdocs.yml
@@ -27,6 +27,8 @@ nav:
- NeMo Relay: integrations/nemo_relay.md
- pi: integrations/pi.md
- Oh My Pi: integrations/oh_my_pi.md
+ - Recipes:
+ - Single-provider coding agents: recipes/single_provider_coding_agents.md
- Routing:
- Overview: routing_algorithms/overview.md
- Task (LLM Classifier): routing_algorithms/llm_classifier_routing.md