langchain-ai/deepagents · error · ValueError

a different CLI provider is already installed for this…

Error message

a different CLI provider is already installed for this process

What it means

get_config_resolver keeps one CLI-tier provider per process in the shared resolver cache. Passing a `cli_provider` that differs from the one already installed raises ValueError: one argv must map to one CLI tier, and silently keeping either provider would misreport the flags the other carries.

Solutions

Reuse the same CLI provider instance for all get_config_resolver calls in the process.
Reset the resolver cache (the module's `_resolver_cache`) between logically distinct phases, e.g. in test setup/teardown.
If flags genuinely changed, restart the process or build a fresh resolver instead of the shared cached one.

Example fix

// before (test)
get_config_resolver(cli_provider=provider_a)
get_config_resolver(cli_provider=provider_b)  # ValueError
// after: reset the shared cache between phases
reset_resolver_cache()  # or fixture that clears _resolver_cache
get_config_resolver(cli_provider=provider_b)

Defensive patterns

Strategy: validation

Validate before calling

def cli_provider_matches_cached(cached_cli, new_cli) -> bool:
    return cached_cli is None or cached_cli == new_cli

assert cli_provider_matches_cached(_resolver_cache.cli_provider, cli_provider), (
    "Process already installed a different CLI provider; reset the cache or reuse it."
)

Try / catch

try:
    resolver = get_config_resolver(cli_provider=new_cli)
except ValueError as exc:
    if "different CLI provider" in str(exc):
        reset_resolver_cache()  # e.g. between test phases / re-parse
        resolver = get_config_resolver(cli_provider=new_cli)
    else:
        raise

Prevention

Parse CLI arguments once per process and reuse the resulting provider for all resolver lookups.
Reset the module-level `_resolver_cache` in test setup/teardown when swapping CLI providers.
Never rebuild an argparse-derived CLI provider mid-process with different flags.

When it happens

Trigger: Calling `get_config_resolver(cli_provider=A)` after a previous call in the same process installed `cli_provider=B` (A != B) — e.g. re-parsing arguments mid-process and rebuilding a different argparse provider, or test code swapping CLI providers without resetting the `_resolver_cache`.

Common situations: Re-running argument parsing twice with different flags in one process (scripts embedding dcode internals); unit tests leaking a cached CLI provider between cases because they didn't reset the module-level cache; programmatic overrides layered on top of real argv.

Related errors

AI-assisted analysis of langchain-ai/deepagents@a1af029e6e (2026-08-29). Data as JSON: /api/errors/516c2027500d3f35. Report an issue: GitHub.

Appendix: source

Thrown at libs/code/deepagents_code/configuration/resolver.py:561

            provider would misreport every flag the other one carries.
    """
    from deepagents_code.configuration.providers import TomlFileProvider
    from deepagents_code.configuration.service import get_managed_snapshot
    from deepagents_code.model_config import DEFAULT_CONFIG_PATH

    if managed_snapshot is not None:
        managed = managed_snapshot
    elif refresh_managed:
        managed = _reload_enforceable_managed_snapshot()
    else:
        managed = get_managed_snapshot()
    key = _ResolverKey(DEFAULT_CONFIG_PATH, managed.status.path)
    with _resolver_cache_lock:
        installed_cli = _resolver_cache.cli_provider
        if cli_provider is not None:
            if installed_cli is not None and installed_cli != cli_provider:
                msg = "a different CLI provider is already installed for this process"
                raise ValueError(msg)
            _resolver_cache.cli_provider = cli_provider
            installed_cli = cli_provider
        entry = _resolver_cache.entry
        if (
            entry is None
            or entry[0] != key
            or (
                cli_provider is not None
                and CLI_RANK not in entry[1].provider_statuses()
            )
        ):
            user_provider = TomlFileProvider(
                name="config.toml", path=DEFAULT_CONFIG_PATH
            )
            user = user_provider.load()
            resolver = resolver_from_snapshots(
                managed=managed,
                user=user,

View on GitHub (pinned to a1af029e6e)