Haufe.ai | Message Validation

Validation rules applied to the messages collections in Chat Completions and Run requests.

The following rules are enforced on message stacks: Either directly provided via /chat/completions or built server-side by attaching messages to a thread.

All violations return an HTTP 422 Unprocessable Entity.

Content Rules

These rules apply to the content field of every message type.

Rule	Details
Non-empty	`content` must not be blank or whitespace-only (applies to `user`, `system`, and `tool` messages)
Maximum length	`content` must not exceed 30,000 characters
Valid Unicode	`content` must be valid UTF-8; broken Unicode characters are rejected

Assistant message content

assistant messages follow a slightly different rule:

content must be non-empty when tool_calls is absent.
content must be empty when tool_calls is present.

Message List Rules

Rule	Details
Non-empty list	The `messages` array must contain at least one message
Last message	The last message must have role `user` or `tool`
System message uniqueness	At most one `system` message is allowed
Assistant message ordering	An `assistant` message must be directly preceded by a `user` message

Example — valid message stack

[
  { "role": "system",    "content": "You are a helpful HR expert." },
  { "role": "user",      "content": "What is BEM?" },
  { "role": "assistant", "content": "BEM stands for ...", "sources": [] },
  { "role": "user",      "content": "Can you elaborate?" }
]

Example — invalid: assistant without preceding user message

[
  { "role": "system",    "content": "You are a helpful HR expert." },
  { "role": "assistant", "content": "BEM stands for ...", "sources": [] }
]

Attachment Rules

Attachments can be added to user messages via the attachments field.

Rule	Details
Maximum attachments	At most 1 attachment per `user` message
Global uniqueness	Each attachment is identified by its `file_id`, `user_id`, and `base_url`. The same attachment must not appear more than once across all messages in a single request

Tool Message Rules

Tool messages (role: "tool") return tool call results to the assistant.

Rule	Details
Must follow an assistant message	A `tool` message must be preceded by an `assistant` message that requested the corresponding tool call
All requested tool calls must be answered	If an `assistant` message requests multiple tool calls, all corresponding `tool` messages must immediately follow it
Unique `tool_call_id`	Each `tool_call_id` must be unique across the entire `messages` array

Example — valid tool message sequence

[
  { "role": "user", "content": "Search for BEM regulations." },
  { "role": "assistant", "content": "", "sources": [], "tool_calls": [
      { "id": "call_1", "type": "function", "function": { "name": "search", "arguments": "{}" } },
      { "id": "call_2", "type": "function", "function": { "name": "fetch", "arguments": "{}" } }
  ]},
  { "role": "tool", "content": "Search result...", "tool_call_id": "call_1" },
  { "role": "tool", "content": "Fetch result...", "tool_call_id": "call_2" }
]

Example — invalid: missing tool message for `call_2`

[
  { "role": "user", "content": "Search for BEM regulations." },
  { "role": "assistant", "content": "", "sources": [], "tool_calls": [
      { "id": "call_1", "type": "function", "function": { "name": "search", "arguments": "{}" } },
      { "id": "call_2", "type": "function", "function": { "name": "fetch", "arguments": "{}" } }
  ]},
  { "role": "tool", "content": "Search result...", "tool_call_id": "call_1" }
]

Tool Choice Rules

When using the tools and tool_choice fields on the request:

Rule	Details
`tool_choice` requires `tools`	`tool_choice` cannot be set if `tools` is absent or empty
Named tool choice must be valid	If `tool_choice` specifies a function by name, that function must be present in the `tools` list

Defaults:

If tools is provided but tool_choice is omitted, it defaults to "auto".
If neither is provided, tool_choice defaults to "none".

Message Validation

On this page