zylon-ai/private-gpt · error · ValueError

Unknown memory type: {type}

Error message

Unknown memory type: {type}

What it means

Raised by Memory.from_defaults when the requested memory type string is not a key in the _PROVIDERS registry. Providers register themselves (presumably via a register decorator writing into _PROVIDERS), and from_defaults only accepts registered types — the Literal['trim', 'summary'] in the signature documents the built-ins. The error is a registry miss: unknown, misspelled, or unregistered type.

Source

Thrown at private_gpt/components/memory/memory.py:34

_PROVIDERS: dict[str, MemoryProvider] = {
    "trim": _trim_memory,
}


def register_memory(memory_type: str, provider: MemoryProvider) -> None:
    _PROVIDERS[memory_type] = provider


class Memory:
    @classmethod
    def from_defaults(
        cls,
        type: Literal["trim", "summary"],
        **kwargs: Any,
    ) -> "BaseMemory":
        provider = _PROVIDERS.get(type)
        if provider is None:
            raise ValueError(f"Unknown memory type: {type}")
        return provider(**kwargs)

View on GitHub (pinned to 4a030776a3)

Solutions

1. Use a registered type string: 'trim' or 'summary' (or your registered custom key).
2. Ensure custom providers are registered by importing their module before calling from_defaults.
3. Check the _PROVIDERS dict keys to see what is actually registered in your runtime.
4. Guard config at load time against the allowed set of memory types.

Example fix

# before
memory = Memory.from_defaults(type='trimm', token_limit=2048)

# after
memory = Memory.from_defaults(type='trim', token_limit=2048)

KNOWN_MEMORY_TYPES = {'trim', 'summary'}
if memory_type not in KNOWN_MEMORY_TYPES:
    raise ConfigError(f'memory type must be one of {KNOWN_MEMORY_TYPES}')

def is_known_memory_type(t: str) -> bool:
    return t in _PROVIDERS

try:
    mem = Memory.from_defaults(type=memory_type, **kw)
except ValueError:
    mem = Memory.from_defaults(type='trim', **kw)

When it happens

Trigger: Memory.from_defaults(type='trimm') (typo); type='summarize' instead of 'summary'; a custom memory provider whose registration code never ran (module not imported, entry point missing); passing a provider class instead of the string key.

Common situations: Typos in memory config; renaming of memory types across versions; custom memory modules not imported before from_defaults is called.

Related errors

• LLM mode '{mode}' is not supported. Available: {available}
• Model '{target_model}' not found. Available: {available}
• Token limit must be set and greater than 0.
• start_on can only be used with 'last' strategy
• include_system can only be used with 'last' strategy

AI-assisted analysis of zylon-ai/private-gpt@4a030776a3 (2026-08-15). Data as JSON: /api/errors/093c72be2f232136. Report an issue: GitHub.