AI Agent (New) 400 error while parallel tool calls

OvertuDece · February 25, 2026, 2:51pm

What is your goal?

I used to use the AI Agent (Old) for my scenario.

Recently, I switch to use AI Agent (New). With the same system prompt and contexts files, it doesn’t work as expected

What is the problem & what have you tried?

Once it tries to query multiple knowledge, it throws below RuntimeError

The AI model which I am using is claude sonnet 4.5 which is distributed by Amazon Bedrock (not Anthropic Connection).

I found there is similar issue indicated that “parallel tool calls” should be the root cause. Currently, make.com not support with this.

This issue occurs for Claude Sonnet and Opus.

I have other scenario using Claude Haiku which working fine.

github.com/agno-agi/agno

[Bug] AwsBedrock fails with "Expected toolResult blocks" when agent makes parallel tool calls

opened 04:56PM - 29 Jan 26 UTC

kerryb48

bug

### Description When using `AwsBedrock`, agents that make **multiple tool calls… in a single response** consistently fail with a `ValidationException` from Bedrock's Converse API. The error indicates that Agno is not properly batching all `toolResult` blocks when sending the next message to Bedrock. This appears related to #6186, but affects direct `AwsBedrock` usage rather than LiteLLM. **Affects all Bedrock models tested:** - Claude Opus 4.5 (`us.anthropic.claude-opus-4-5-20251101-v1:0`) - Claude Sonnet 3.5 (`us.anthropic.claude-3-5-sonnet-20241022-v2:0`) - Amazon Nova Pro (`amazon.nova-pro-v1:0`) ### Steps to Reproduce 1. Create an agent with `AwsBedrock` model and a toolkit with multiple tools 2. Send a query that causes the LLM to call multiple tools in parallel 3. Agent fails with `ValidationException` ### Agent Configuration (if applicable) **Minimal reproduction:** ```python import asyncio from agno.agent import Agent from agno.models.aws import AwsBedrock from agno.tools.calculator import CalculatorTools agent = Agent( id="calc-agent", name="Calculator Agent", model=AwsBedrock( id="us.anthropic.claude-opus-4-5-20251101-v1:0", aws_region="us-east-1" ), tools=[CalculatorTools()], instructions=["Perform all requested calculations."], ) async def test(): # This query triggers 2 parallel tool calls (add + subtract) response = await agent.arun("Calculate 25 + 17 AND 100 - 37") print(response.content) asyncio.run(test()) ``` ### Expected Behavior Agent successfully executes multiple tool calls and returns a combined response without error. ### Actual Behavior Agent fails with: ``` ValidationException: Expected toolResult blocks at messages.2.content for the following Ids: tooluse_xxx, tooluse_yyy ``` ### Screenshots or Logs (if applicable) **Tested across multiple agents and toolkits:** | Agent | Tools Available | Parallel Query | Success Rate | |-------|----------------|----------------|--------------| | Finance (YFinance) | 9 tools | Price + fundamentals + info | **0/10** | | HackerNews | 2 tools | Stories + user details | **0/3** | | DuckDuckGo | 2 tools | Web search + news search | **0/3** | | Calculator | 8 tools | Multiple calculations | **0/3** | **Serial tool calls work correctly:** | Test | Pattern | Success Rate | |------|---------|--------------| | Single tool call | 1 tool per response | **5/5** | | Serial (instructed) | 1 tool at a time | **5/5** | --- Captured by patching `_format_messages()`, here's the actual message structure Agno sends to Bedrock: ``` Query: Calculate 25 + 17 AND 100 - 37 FORMATTED MESSAGES SENT TO BEDROCK: --- Message 0 --- Role: user Content blocks: 1 Content[0]: ['text'] --- Message 1 --- Role: assistant Content blocks: 2 Content[0]: ['toolUse'] toolUseId: tooluse_-ZKveBMLRrqqVIO43oBqHQ ← add name: add Content[1]: ['toolUse'] toolUseId: tooluse_DR_3f80OR0aRzl1SfLr3yw ← subtract name: subtract --- Message 2 --- Role: user Content blocks: 1 Content[0]: ['toolResult'] toolUseId: tooluse_-ZKveBMLRrqqVIO43oBqHQ ← Only "add" result! --- Message 3 --- Role: user Content blocks: 1 Content[0]: ['toolResult'] toolUseId: tooluse_DR_3f80OR0aRzl1SfLr3yw ← "subtract" result in SEPARATE message! ERROR: ValidationException: Expected toolResult blocks at messages.2.content for the following Ids: tooluse_DR_3f80OR0aRzl1SfLr3yw ``` ### Environment ```markdown - OS: macOS 24.6.0 (darwin) - Python: 3.12+ - agno: 2.4.7 (also tested on 2.4.6) - Model: us.anthropic.claude-opus-4-5-20251101-v1:0 (Bedrock) - AWS Region: us-east-1 ``` ### Possible Solutions (optional) **The bug appears to be in `_format_messages()` in `agno/models/aws/bedrock.py` (lines 261-268):** ```python elif message.role == "tool": content = message.get_content(use_compressed_content=compress_tool_results) tool_result = { "toolUseId": message.tool_call_id, "content": [{"json": {"result": content}}], } formatted_message: Dict[str, Any] = {"role": "user", "content": [{"toolResult": tool_result}]} formatted_messages.append(formatted_message) # ← Creates SEPARATE message for each tool result! ``` Each tool result message creates a **separate user message**. But Bedrock requires all `toolResult` blocks to be in a **single user message** following the assistant's `toolUse` response. **What Agno sends:** ``` Message 1: assistant with [toolUse_1, toolUse_2] Message 2: user with [toolResult_1] ← WRONG: separate message Message 3: user with [toolResult_2] ← WRONG: separate message ``` **What Bedrock expects:** ``` Message 1: assistant with [toolUse_1, toolUse_2] Message 2: user with [toolResult_1, toolResult_2] ← CORRECT: single message with both ``` ### Additional Context Bug is **100% reproducible** when LLM makes 2+ tool calls in a single response - Bug affects **all toolkits** tested, not specific to YFinance - Bug does **not occur** with single tool calls - Bug affects **all Bedrock models** tested: - Claude Opus 4.5: 0/10 success - Claude Sonnet 3.5: 0/10 success - Amazon Nova Pro: 0/6 success ## References - [Parallel Tool Execution with Bedrock Converse API](https://medium.com/@gor17v/parallel-tool-execution-with-claude-4-building-high-performing-agents-with-bedrock-converse-api-433da0efab60) - Shows correct pattern: "CRITICAL: Add ALL tool results to a SINGLE user message" ## Workaround Instructing the agent to perform tool calls sequentially seems to generally work.

github.com/agno-agi/agno

[fix] align toolResult/toolUse blocks for LiteLLM Bedrock path (#6635)

main ← hztBUAA:fix/litellm-bedrock-tool-mismatch

opened 09:14PM - 19 Feb 26 UTC

hztBUAA

+474 -15

## Summary - Align toolResult/toolUse message blocks when using LiteLLM with Be…drock provider - Batch consecutive tool result messages into a single user message for the Bedrock API contract - Ensure tool_use_id references match between assistant tool_use and user tool_result blocks Fixes #6186 ## Type of change - [x] Bug fix (non-breaking change which fixes an issue) ## Checklist - [x] My code follows the style guidelines of this project (`./scripts/format.sh` and `./scripts/validate.sh` pass) - [x] I have performed a self-review of my code - [x] New unit tests added - [x] Existing tests still pass ## Additional Notes **Root cause:** When using LiteLLM with a Bedrock backend, tool result messages were sent as separate user messages rather than batched together. The Bedrock/Anthropic API requires all tool results from parallel tool calls to be in a single user message. Additionally, tool_use_id references could become misaligned between the assistant's tool_use blocks and the user's tool_result blocks. **Fix:** Added batching logic to the LiteLLM chat module to merge consecutive tool result messages and ensure proper ID alignment.

github.com/anthropics/claude-code

API Error 400: tool_use blocks without corresponding tool_result blocks

opened 07:11AM - 13 Aug 25 UTC

closed 08:36PM - 18 Sep 25 UTC

InfoCodeStreamly

bug duplicate platform:macos area:core area:api

## Environment - Platform (select one): - [x] Anthropic API - [ ] AWS …Bedrock - [ ] Google Vertex AI - [ ] Other: - Claude CLI version: 1.0.77 (Claude Code) - Operating System: macOS (Darwin 24.6.0) - Terminal: Terminal App ## Bug Description Frequent API Error 400 with message "tool_use ids were found without tool_result blocks immediately after". This prevents normal operation and requires retrying commands. ## Steps to Reproduce 1. Use Claude Code CLI for any task involving tool calls 2. Error occurs randomly during tool execution 3. Specific error: `messages.129: tool_use ids were found without tool_result blocks immediately after: toolu_01JX7gsswzmXeYhXg6F8wGRy` ## Expected Behavior Tool calls should complete successfully with proper tool_use/tool_result pairing. ## Actual Behavior API returns 400 error indicating missing tool_result blocks, interrupting workflow. Issue has become frequent over the past couple of days. ## Additional Context - Node.js version: v24.2.0 (getting npm warnings about engine compatibility) - Just updated Claude Code CLI via `npm update -g @anthropic/claude-code` - Retrying the same command usually works on second attempt - Error seems to happen with various tool types, not specific to one tool API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.129: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01JX7gsswzmXeYhXg6F8wGRy. Each `tool_use` block must have a corresponding `tool_result` block in the next message."}}

Error messages or input/output bundles

Error Sample:
400 | Expected toolResult blocks at messages.2.content for the following Ids:

tooluse_NEAYa0o06upj2NH8TGnla9, tooluse_zpyCTOewDT11rXFDEeFWLk

Topic		Replies	Views
Make AI Agent (New) 500 issue Questions error , ai , ai-agent	5	46	February 9, 2026
Need Help in Connecting AWS Bedrock AI Agent to Make Questions ai	2	84	May 8, 2025
AI Agent - tools do not run Questions connections , error , ai-agent	3	70	March 11, 2026
500 \| An error occurred in AI Agent Service Questions ai , ai-agent	4	32	February 19, 2026
Ai Agent Interrupted by Customer/User Questions ai , ai-agent	0	18	January 19, 2026

AI Agent (New) 400 error while parallel tool calls

What is your goal?

What is the problem & what have you tried?

Error messages or input/output bundles

Screenshots (scenario flow, module settings, errors)

Related topics