Bedrock — Amazon Bedrock Runtime

AWS docs: https://docs.aws.amazon.com/bedrock/latest/APIReference/

Amazon Bedrock Runtime uses the REST JSON protocol, served at the paths AWS binds it to: POST /model/{modelId}/invoke and POST /model/{modelId}/converse. No model is invoked — both answer from a canned response.

---

Notes

• modelId may be a model ID, an inference-profile ID, or an ARN. It is a single path segment, so an SDK percent-encodes an ARN’s separators; the route matches either form.
• Converse returns a complete ConverseResponse — every member the AWS model marks required, with the canned assistant text in output.message.content[0].text. Because nothing is inferred, the token counts and metrics.latencyMs are all zero.
• InvokeModel takes and returns an opaque payload whose format belongs to the model, not to Bedrock. Overcast emulates no model, so it answers with a single overcastEmulator field rather than imitating one family’s shape and being wrong for the rest. Use Converse where you need a response shape AWS itself defines.
• The streaming and token-counting siblings (InvokeModelWithResponseStream, ConverseStream, CountTokens) are not emulated and return a JSON 501 Not Implemented, as does any other unimplemented operation.

Summary

Category	🧊 Inert
Inference	2

---

Endpoints

Inference

Operation	Status	Notes	AWS Docs
InvokeModel	🧊 Inert	POST /model/{modelId}/invoke — accepted and answered 200; the required payload is not parsed and no model runs. The response body is an opaque blob whose format belongs to the model, which Overcast does not emulate, so it returns one self-describing field rather than any model family’s shape	docs
Converse	🧊 Inert	POST /model/{modelId}/converse — accepted and answered with a complete ConverseResponse, every @required member present; no inference runs, so the assistant text is canned and the token counts and latency are zero	docs