OpenBMB/ChatDev · error · ConfigError

model.input_mode must be 'prompt' or 'messages'

Error message

model.input_mode must be 'prompt' or 'messages'

What it means

The optional 'input_mode' field of an agent's model config could not be mapped to the AgentInputMode enum. Only 'prompt' (single string input) and 'messages' (structured chat messages, the default) are accepted; matching is case-insensitive after stripping.

Source

Thrown at entity/configs/node/agent.py:361

        mapping = require_mapping(data, path)
        provider = require_str(mapping, "provider", path)
        base_url = optional_str(mapping, "base_url", path)
        name_value = mapping.get("name")
        if isinstance(name_value, str) and name_value.strip():
            model_name = name_value.strip()
        else:
            raise ConfigError("model.name must be a non-empty string", extend_path(path, "name"))

        role = optional_str(mapping, "role", path)
        api_key = optional_str(mapping, "api_key", path)
        params = optional_dict(mapping, "params", path) or {}
        raw_input_mode = optional_str(mapping, "input_mode", path)
        input_mode = AgentInputMode.MESSAGES
        if raw_input_mode:
            try:
                input_mode = AgentInputMode(raw_input_mode.strip().lower())
            except ValueError as exc:
                raise ConfigError(
                    "model.input_mode must be 'prompt' or 'messages'",
                    extend_path(path, "input_mode"),
                ) from exc

        tooling_cfg: List[ToolingConfig] = []
        if "tooling" in mapping and mapping["tooling"] is not None:
            raw_tooling = mapping["tooling"]
            if not isinstance(raw_tooling, list):
                 raise ConfigError("tooling must be a list", extend_path(path, "tooling"))
            for idx, item in enumerate(raw_tooling):
                tooling_cfg.append(
                    ToolingConfig.from_dict(item, path=extend_path(path, f"tooling[{idx}]"))
                )

        thinking_cfg = None
        if "thinking" in mapping and mapping["thinking"] is not None:
            thinking_cfg = ThinkingConfig.from_dict(mapping["thinking"], path=extend_path(path, "thinking"))

View on GitHub (pinned to 4fb2db0ea9)

Solutions

1. Set input_mode to exactly 'prompt' or 'messages' (any casing works since it is lowered)
2. Omit input_mode entirely if you want the default 'messages' behavior
3. Search your config source for the bad literal using the error's path

Example fix

// before
{"input_mode": "chat"}
// after
{"input_mode": "prompt"}

ALLOWED_MODES = {'prompt', 'messages'}
mode = (cfg.get('model', {}).get('input_mode') or 'messages').strip().lower()
if mode not in ALLOWED_MODES:
    cfg['model']['input_mode'] = 'messages'

def valid_input_mode(cfg: dict) -> bool:
    m = cfg.get('model', {}).get('input_mode')
    return m is None or (isinstance(m, str) and m.strip().lower() in {'prompt', 'messages'})

try:
    ModelConfig.from_dict(data, path='agent')
except ConfigError as e:
    if 'input_mode' in e.path:
        data['model'].pop('input_mode', None)  # fall back to default
        ModelConfig.from_dict(data, path='agent')
    else:
        raise

When it happens

Trigger: Setting model.input_mode to anything other than 'prompt' or 'messages' (case-insensitive), e.g. 'chat', 'text', 'completion', or a typo like 'promt'. Absent/null input_mode is fine and defaults to MESSAGES.

Common situations: Copy-pasting configs from other agent frameworks that use 'chat'/'text' modes; typos; version changes where old mode names were removed.

Understand the failure class

Background: Invalid enum value errors: "Unknown type", "Invalid scope", "must be one of" — when a string is not on the library's allowed list — this error's family across 23 libraries.

Related errors

• model.name must be a non-empty string
• tooling must be a list
• memories must be a list
• role must be 'user' or 'assistant'
• duration_unit must be one of: {', '.join(valid_units)}

AI-assisted analysis of OpenBMB/ChatDev@4fb2db0ea9 (2026-08-27). Data as JSON: /api/errors/db837bf42c2e38ca. Report an issue: GitHub.