feat: AWS Bedrock provider adapter

README.md

@@ -87,8 +87,9 @@ Wizard defaults reflect the project's per-phase recommendations (stronger reason
 | `anthropic` | [console.anthropic.com](https://console.anthropic.com/settings/keys) | Reference adapter. NOT included in Claude Pro / Max subscriptions — separate billing. |
 | `openai` | [platform.openai.com](https://platform.openai.com/api-keys) | NOT included in ChatGPT / Codex subscriptions — separate billing. |
 | `google` | [aistudio.google.com](https://aistudio.google.com/apikey) | NOT included in Gemini Advanced — separate billing. |
+| `bedrock` | — (AWS credential chain) | Claude on AWS Bedrock. No `api_key`: credentials come from `AWS_ACCESS_KEY_ID`/`AWS_SECRET_ACCESS_KEY` env vars or a `~/.aws` profile, region from `AWS_REGION`. Model IDs are inference profiles (`us.anthropic.claude-sonnet-4-6`, `global.anthropic.claude-haiku-4-5-20251001-v1:0`, ...) — enable them under "Model access" in the Bedrock console and list them with `aws bedrock list-inference-profiles`. Not offered by the setup wizard yet; configure by hand — full guide: [`utilities/llm/providers/BEDROCK.md`](libs/openant-core/utilities/llm/providers/BEDROCK.md). |
 
-All three support tool calling, so any of them can drive the `enhance` and `verify` phases that use the agentic tool-use loop.
+All four support tool calling, so any of them can drive the `enhance` and `verify` phases that use the agentic tool-use loop.
 
 #### Quick path for Anthropic-only setups
 
@@ -216,7 +217,7 @@ openant project switch <org/repo> # switch active project
 
 Things on the list, in no particular order:
 
-- **More provider adapters.** Ollama (local models), vLLM, Cohere, Mistral, Groq, Amazon Bedrock, Azure OpenAI — each is a small Python adapter recipe (plus a few Go wizard/probe touch-points if you want it offered by `openant setup llm`) per the contributor guide. Lower the barrier to local / on-prem inference.
+- **More provider adapters.** Ollama (local models), vLLM, Cohere, Mistral, Groq, Azure OpenAI — each is a small Python adapter recipe (plus a few Go wizard/probe touch-points if you want it offered by `openant setup llm`) per the contributor guide. Lower the barrier to local / on-prem inference.
 - **Subscription-based auth.** ChatGPT / Codex, Claude Pro / Max, and Gemini Advanced subscriptions don't currently grant API quota — users have to maintain a separate API-tier key per provider. OAuth-based adapters that ride the consumer subscription would close that gap.
 - **Cross-provider tool-call quirks.** All three shipped adapters support tool calling, but the long tail (parallel tool calls, strict-mode schema enforcement, retry semantics on partial JSON) behaves differently per provider. Real-world scans surface these — PRs welcome.
 - **More languages.** The supported-languages list above is current coverage. Rust, Java, C#, and Swift come up frequently.

config/models.json

@@ -38,6 +38,42 @@
       "source": "liveness contested and not verifiable in build env; asserted neither alive nor dead",
       "retrieved": "2026-07-23"
     },
+    {
+      "id": "us.anthropic.claude-opus-4-8", "provider": "bedrock", "status": "current",
+      "price": {"input": 15.00, "output": 75.00},
+      "source": "verified against a live Bedrock account via list-inference-profiles (us-east-1, status ACTIVE); price mirrors the anthropic claude-opus-4-8 record (Bedrock Claude token rates match the direct Anthropic API)",
+      "retrieved": "2026-07-31"
+    },
+    {
+      "id": "global.anthropic.claude-opus-4-8", "provider": "bedrock", "status": "current",
+      "price": {"input": 15.00, "output": 75.00},
+      "source": "verified against a live Bedrock account via list-inference-profiles (us-east-1, status ACTIVE); price mirrors the anthropic claude-opus-4-8 record (Bedrock Claude token rates match the direct Anthropic API)",
+      "retrieved": "2026-07-31"
+    },
+    {
+      "id": "us.anthropic.claude-sonnet-4-6", "provider": "bedrock", "status": "current",
+      "price": {"input": 3.00, "output": 15.00},
+      "source": "verified against a live Bedrock account via list-inference-profiles (us-east-1, status ACTIVE); price mirrors the anthropic claude-sonnet-4-6 record (Bedrock Claude token rates match the direct Anthropic API)",
+      "retrieved": "2026-07-31"
+    },
+    {
+      "id": "global.anthropic.claude-sonnet-4-6", "provider": "bedrock", "status": "current",
+      "price": {"input": 3.00, "output": 15.00},
+      "source": "verified against a live Bedrock account via list-inference-profiles (us-east-1, status ACTIVE); price mirrors the anthropic claude-sonnet-4-6 record (Bedrock Claude token rates match the direct Anthropic API)",
+      "retrieved": "2026-07-31"
+    },
+    {
+      "id": "us.anthropic.claude-haiku-4-5-20251001-v1:0", "provider": "bedrock", "status": "current",
+      "price": {"input": 1.00, "output": 5.00},
+      "source": "verified against a live Bedrock account via list-inference-profiles (us-east-1, status ACTIVE); price mirrors the anthropic claude-haiku-4-5-20251001 record (Bedrock Claude token rates match the direct Anthropic API)",
+      "retrieved": "2026-07-31"
+    },
+    {
+      "id": "global.anthropic.claude-haiku-4-5-20251001-v1:0", "provider": "bedrock", "status": "current",
+      "price": {"input": 1.00, "output": 5.00},
+      "source": "verified against a live Bedrock account via list-inference-profiles (us-east-1, status ACTIVE); price mirrors the anthropic claude-haiku-4-5-20251001 record (Bedrock Claude token rates match the direct Anthropic API)",
+      "retrieved": "2026-07-31"
+    },
     {
       "id": "gpt-4o", "provider": "openai", "status": "current",
       "price": {"input": 2.50, "output": 10.00},

libs/openant-core/core/model_registry.py

@@ -39,7 +39,7 @@
 _CONFIG_REL = Path("config") / "models.json"
 _SEARCH_LEVELS = 6
 _VALID_STATUS = frozenset({"current", "retired", "unknown"})
-_VALID_PROVIDERS = frozenset({"anthropic", "openai", "google"})
+_VALID_PROVIDERS = frozenset({"anthropic", "openai", "google", "bedrock"})
 
 
 def _search_upward(start: Path) -> Path | None:

libs/openant-core/tests/_llm_factories/bedrock.py

@@ -0,0 +1,152 @@
+"""Scenario factory for the Bedrock adapter contract tests.
+
+Each scenario builds a fake ``anthropic.AnthropicBedrock`` client wired
+with the right scripted behavior, then constructs a
+:class:`BedrockAdapter` over that fake. Bedrock speaks the same
+Messages API through the same ``anthropic`` SDK types and exception
+classes as the direct API, so the scripted responses and raised
+exceptions mirror the Anthropic factory — only the client spec and the
+endpoint URL differ.
+
+See ``tests/test_llm_adapter_contract.py`` for the scenario catalogue.
+"""
+
+from __future__ import annotations
+
+from types import SimpleNamespace
+from typing import Any
+from unittest.mock import MagicMock
+
+import anthropic
+import httpx
+
+from utilities.llm import LLMAdapter
+from utilities.llm.providers.bedrock import BedrockAdapter
+
+
+_ENDPOINT = "https://bedrock-runtime.us-east-1.amazonaws.com/model/test/invoke"
+
+
+def _text_block(text: str) -> SimpleNamespace:
+    return SimpleNamespace(type="text", text=text)
+
+
+def _tool_use_block(*, id: str, name: str, input: dict) -> SimpleNamespace:
+    return SimpleNamespace(type="tool_use", id=id, name=name, input=input)
+
+
+def _response(
+    *, content: list, input_tokens: int, output_tokens: int, stop_reason: str
+) -> SimpleNamespace:
+    return SimpleNamespace(
+        content=content,
+        usage=SimpleNamespace(
+            input_tokens=input_tokens, output_tokens=output_tokens
+        ),
+        stop_reason=stop_reason,
+    )
+
+
+def _fake_httpx_response(status_code: int, *, retry_after: str | None = None) -> httpx.Response:
+    headers = {}
+    if retry_after is not None:
+        headers["retry-after"] = retry_after
+    return httpx.Response(
+        status_code=status_code,
+        headers=headers,
+        request=httpx.Request("POST", _ENDPOINT),
+    )
+
+
+def _script_text(call_args: dict) -> SimpleNamespace:
+    # The contract test asserts content=="hi there", usage 3/5, end_turn.
+    return _response(
+        content=[_text_block("hi there")],
+        input_tokens=3,
+        output_tokens=5,
+        stop_reason="end_turn",
+    )
+
+
+def _script_tool_use_round(call_args: dict) -> SimpleNamespace:
+    has_assistant = any(m.get("role") == "assistant" for m in call_args["messages"])
+    if not has_assistant:
+        return _response(
+            content=[
+                _tool_use_block(
+                    id="toolu_test_1",
+                    name="echo",
+                    input={"text": "hello"},
+                )
+            ],
+            input_tokens=10,
+            output_tokens=8,
+            stop_reason="tool_use",
+        )
+    return _response(
+        content=[_text_block("echoed: hello")],
+        input_tokens=20,
+        output_tokens=4,
+        stop_reason="end_turn",
+    )
+
+
+def _raise_auth(_call_args: dict):
+    raise anthropic.AuthenticationError(
+        message="invalid signature",
+        response=_fake_httpx_response(401),
+        body=None,
+    )
+
+
+def _raise_rate_limit(_call_args: dict):
+    # Bedrock throttling (ThrottlingException) surfaces as the SDK's
+    # standard 429 RateLimitError.
+    raise anthropic.RateLimitError(
+        message="too many requests, please wait before trying again",
+        response=_fake_httpx_response(429, retry_after="7"),
+        body=None,
+    )
+
+
+def _raise_connection(_call_args: dict):
+    raise anthropic.APIConnectionError(
+        request=httpx.Request("POST", _ENDPOINT),
+    )
+
+
+def _raise_not_found(_call_args: dict):
+    raise anthropic.NotFoundError(
+        message="model not found: ghost-model",
+        response=_fake_httpx_response(404),
+        body=None,
+    )
+
+
+_SCENARIO_HANDLERS = {
+    "text": _script_text,
+    "tool_use_round": _script_tool_use_round,
+    "auth_error": _raise_auth,
+    "rate_limit": _raise_rate_limit,
+    "connection_error": _raise_connection,
+    "model_not_found": _raise_not_found,
+    "validate_ok": _script_text,        # any valid response satisfies validate
+    "validate_auth_fail": _raise_auth,  # validate is a thin wrapper over create
+}
+
+
+def make_adapter(scenario: str) -> LLMAdapter:
+    """Build a BedrockAdapter whose SDK is scripted for ``scenario``."""
+    if scenario not in _SCENARIO_HANDLERS:
+        raise KeyError(f"Unknown scenario: {scenario!r}")
+
+    handler = _SCENARIO_HANDLERS[scenario]
+
+    def side_effect(**kwargs: Any) -> Any:
+        return handler(kwargs)
+
+    fake_client = MagicMock(spec=anthropic.AnthropicBedrock)
+    fake_client.messages = MagicMock()
+    fake_client.messages.create = MagicMock(side_effect=side_effect)
+
+    return BedrockAdapter(_client=fake_client)

libs/openant-core/tests/test_llm_adapter_contract.py

@@ -112,12 +112,19 @@ def _google_factory():
     return make_adapter
 
 
+def _bedrock_factory():
+    from tests._llm_factories.bedrock import make_adapter
+
+    return make_adapter
+
+
 # Each row: (display_name, scenario_factory_callable)
 # Add a row when registering a new adapter.
 ADAPTERS: list[tuple[str, Callable[[str], LLMAdapter]]] = [
     ("anthropic", _anthropic_factory()),
     ("openai", _openai_factory()),
     ("google", _google_factory()),
+    ("bedrock", _bedrock_factory()),
 ]