Cloud agent platforms

Bonito vs Amazon Bedrock AgentCore

AWS's agent runtime vs. agents and failover across every cloud

Amazon Bedrock AgentCore is AWS's managed set of agent services (Runtime, Harness, Gateway, Memory, Identity, Policy, Observability, Evaluations) for agents built in any framework, with Bedrock as the default model source. Since July 2026 its Gateway also fronts OpenAI and Anthropic directly. Bonito adds automatic cross-provider failover, cost routing, Azure and Vertex, and dollar cost per agent on top of your own accounts.

AWS AgentCore pricing

Pay per use per component, no platform fee: Runtime $0.1276/vCPU-hr, Gateway $0.005 per 1K invocations, Memory $0.25 per 1K short-term events. Model tokens billed separately at Bedrock rates.

Model costs

AWS sells Bedrock models at its per-token rates. Gateway targets for OpenAI or Anthropic use your own keys, plus the per-invocation gateway fee.

Our benchmark

We ran the same agent swarm on both

15 support questions, three rounds, every number published. AgentCore's default stack used 16.4 million tokens and $11.87, and hallucinated a refund policy. Bonito cost 3 cents, declined honestly on every trap, kept answering through a provider outage, and ran one swarm across four clouds.

Read the benchmark and the receipts

What AWS AgentCore does well

  • Framework-agnostic hosting for LangChain, OpenAI Agents SDK, Claude Agent SDK, Strands or your own code
  • Config-first harness with model switching mid-session and a CLI (agentcore deploy) plus CDK constructs
  • Gateway turns APIs, Lambda and OpenAPI specs into MCP tools, with one-click Slack, Salesforce, Jira and Zendesk
  • Deep compliance: HIPAA-eligible, FedRAMP, SOC 1/2/3, ISO 27001
  • Evaluations, batch evals and A/B tests on live traffic

Where Bonito goes further

  • Automatic failover to an equivalent model on another provider; AgentCore spreads load across matching targets but documents no retry-on-failure
  • Azure and Vertex alongside Bedrock and the direct providers, eight providers in all, on your own accounts with 0% markup
  • Cost routing to cheaper models for simple work, across providers
  • Dollar budgets per agent and dollar cost per request, agent, team and provider; AgentCore limits tokens and iterations
  • Built-in approval queue and cron schedules, instead of wiring Step Functions
  • Data residency pinning to CA, US or EU across every provider in one setting

Where AWS AgentCore is ahead of Bonito

  • Compliance, including FedRAMP
  • Managed browser and code interpreter tools
  • Evaluations and live A/B testing of agents
  • Knowledge base connectors for SharePoint, Confluence, Google Drive and OneDrive

Capability by capability

CapabilityBonitoAWS AgentCore
Agents
Build and run agentsYesYes
Approval queue and scheduled runsYesPartialStep Functions for schedules and approval callbacks
Dollar budgets per agentYesPartialiteration, token and rate limits, not dollars
MCP supportYesYes
Routing
OpenAI-compatible gatewayYesYes
Automatic failover across providersYesNonot found in their docs (Sep 2026); matching targets are load-spread
Cheaper models for simple work (cost routing)YesPartialBedrock Intelligent Prompt Routing, within one model family
Bedrock, Vertex, Azure and direct providers in one placeYesPartialBedrock, OpenAI, Anthropic and OpenAI-compatible endpoints; no Azure or Vertex connector
Semantic cachingNoNonot found in their docs (Sep 2026); Bedrock has prompt caching
Governance
SSO, RBAC and audit logYesYes
Data residency pinningYesYes
Observability and knowledge
Agent traces with cost per request and agentYesPartialtoken metrics in CloudWatch; no per-agent dollar view found
Knowledge bases (RAG)YesYes
How you run it
Whole stack from one config fileYesYesharness.json and agentcore deploy
Self-hosted or open sourcePartialPartialSDKs and CLI are open source; services are AWS only
No markup on inferenceYesPartialAWS resells Bedrock models; BYO keys for OpenAI and Anthropic targets

Checked against AWS AgentCore's own documentation on 2026-09-27. Products change quickly; if something here is out of date, tell us at hello@trybonito.com and we will fix it.

Sources

See Bonito on your own stack

Start free, or get a report on where agents fit in your company first.