AWS Bedrock

Access Claude, Nova, Llama, and more through AWS infrastructure with enterprise-grade security and compliance.

Prerequisites

• AWS account with Bedrock enabled in your region
• Model access granted in the Bedrock Console (some models require approval)
• AWS credentials configured (see authentication below)

Configuration

models:
  bedrock-claude:
    provider: amazon-bedrock
    model: global.anthropic.claude-sonnet-5
    max_tokens: 64000
    provider_opts:
      region: us-east-1

Authentication

Option 1: Bedrock API Key (Simplest)

export AWS_BEARER_TOKEN_BEDROCK="your-key"

models:
  bedrock:
    provider: amazon-bedrock
    model: global.anthropic.claude-sonnet-5
    token_key: AWS_BEARER_TOKEN_BEDROCK # env var name
    provider_opts:
      region: us-east-1

Option 2: AWS Credentials (Default)

Uses the standard AWS SDK credential chain: env vars → shared credentials → config → IAM roles.

models:
  bedrock:
    provider: amazon-bedrock
    model: global.anthropic.claude-sonnet-5
    provider_opts:
      profile: my-aws-profile
      region: us-east-1

With IAM Role Assumption

models:
  bedrock:
    provider: amazon-bedrock
    model: global.anthropic.claude-sonnet-5
    provider_opts:
      role_arn: "arn:aws:iam::123456789012:role/BedrockAccessRole"
      external_id: "my-external-id"

Provider Options

Inference Profiles

Inference profiles route requests across the regions supported by that profile:

Claude on Bedrock: transport and limits

Docker Agent calls ConverseStream on Bedrock runtime. Native Mantle, InvokeModel, Bedrock Messages and Claude Platform on AWS are not wired by this integration. A custom endpoint does not switch the runtime client's API.

Mythos 5 is Mantle-only and explicitly unsupported on runtime/Converse. Mythos 5.1 is documented on runtime, with gated access, not Mantle. Vendor support does not mean this client's complete 5.x compatibility work has shipped.

AWS compaction requires InvokeModel for both compact-2026-01-12 and compact-2026-09-04; neither works through Converse. Haiku 5.5, Opus 5/5.5 and Sonnet 5/5.5 cards document the latter; Fable 5/5.1 and Mythos 5.1 cards do not document compaction. The current client also does not wire Claude API cache diagnostics, server-side fallbacks or native JSON-schema structured output. General Bedrock feature statements do not override a specific model card.

See Claude access paths for the dated nine-model evidence matrix, exact card-sourced IDs, structured-output limits and account prerequisites. apac. in the general profile table is not a documented Claude 5.x profile in this audit; do not generate IDs by copying prefixes.

Thinking Budget (Claude on Bedrock)

Bedrock Claude models support extended thinking — an internal reasoning phase before the model produces its response. Set thinking_budget to a token count (1024–32768) or an effort level string that maps automatically:

models:
  bedrock-claude-thinking:
    provider: amazon-bedrock
    model: global.anthropic.claude-sonnet-5
    thinking_budget: 8192   # tokens, or use an effort string like "medium"
    max_tokens: 16384       # must be > thinking_budget
    provider_opts:
      region: us-east-1

thinking_budget must be ≥ 1024 and less than max_tokens. Values outside this range are logged as a warning and ignored.

Adaptive thinking (Opus 4.6+)

Newer Claude Opus models (4.6, 4.7, 4.8) reject token-based thinking — Bedrock returns a ValidationException asking for thinking.type=adaptive. For these models, use adaptive thinking:

models:
  bedrock-opus-adaptive:
    provider: amazon-bedrock
    model: global.anthropic.claude-opus-4-8
    thinking_budget: adaptive/high   # adaptive | adaptive/low | adaptive/medium | adaptive/high | adaptive/xhigh | adaptive/max
    provider_opts:
      region: us-east-1

Docker Agent recognizes these models (including Bedrock-style IDs) and transparently coerces a configured token budget or effort level to adaptive thinking, logging a warning — so thinking_budget: 32768 on Opus 4.8 won't fail, but adaptive or adaptive/<effort> is the recommended configuration. On older models that still use token-based thinking (e.g. Sonnet 4.5), adaptive is forwarded as-is and rejected by Bedrock — use a token count or effort level there instead.

Claude 5.x models

The nine Claude 5.x models listed in the Anthropic provider guide follow the same rules on Bedrock, including Bedrock-style IDs such as us.anthropic.claude-sonnet-5-5-v1:0:

• Effort levels and adaptive are sent as adaptive thinking. An integer token budget becomes adaptive thinking with the model's default effort.
• temperature, top_p and top_k are never sent.
• thinking_budget: none sends thinking: {"type": "disabled"} on Haiku 5.5, Opus 5 and Sonnet 5. The other six get adaptive thinking with low effort and a startup warning.
• Reasoning blocks are stored and replayed exactly as Bedrock returned them, in order: signed thinking (including signature-only thinking, the default display on these models), redacted thinking and the text and tool calls around them. If a message changed after it was received but its reasoning did not (for example a sanitized tool name or a redacted tool argument), its recorded reasoning is still replayed in its original place, with the current text and tool calls substituted. If the reasoning itself changed, it is sent without it, with a warning in the logs; inside the tool loop the next request continues, with thinking on, the request is refused instead, and a new user turn starts a fresh loop. Sessions saved by older versions keep that flattened replay; older versions ignore the stored blocks.
• Mythos 5 is refused at startup: AWS serves it only through Bedrock Mantle, which Docker Agent does not support. Use Mythos 5.1, which is available on Converse, or the Claude API.

Interleaved Thinking (Claude on Bedrock)

Interleaved thinking lets the model reason between tool calls, not just at the start. This is useful for complex agentic tasks. Enable it alongside a thinking budget:

models:
  bedrock-claude-interleaved:
    provider: amazon-bedrock
    model: global.anthropic.claude-sonnet-5
    thinking_budget: high
    provider_opts:
      region: us-east-1
      # interleaved_thinking is auto-enabled when thinking_budget is set

Docker Agent auto-enables interleaved_thinking whenever a thinking budget is configured on a Bedrock-hosted Claude model and automatically adds the interleaved-thinking-2025-05-14 beta header. If you set interleaved_thinking: false while a thinking budget is active, a warning is logged because the budget may be ignored by Bedrock without the beta header.

See the Thinking / Reasoning guide for a full cross-provider overview.

Prompt Caching

Automatically enabled for supported models to reduce latency and costs. System prompts, tool definitions, and recent messages are cached with a 5-minute TTL.

# Disable if needed
provider_opts:
  disable_prompt_caching: true