Graphify-Labs/graphify · error · ImportError

the 'anthropic' package is required for this backend but is

Error message

the 'anthropic' package is required for this backend but is not installed. Install it with:  uv tool install "graphifyy[anthropic]" --force  (uv tool), or  pip install anthropic  (pip/venv install).

What it means

Raised by `_call_claude` when `import anthropic` fails. The native `claude` backend (not the OpenAI compat layer, not claude-cli) calls Anthropic's SDK directly, so the `anthropic` package must be present. The message comes from `_backend_pkg_hint("anthropic", "anthropic")` and includes the exact uv/pip install command.

Source

Thrown at graphify/llm.py:1320

    if output_tokens < 50 and backend == "ollama":
        print(
            "[graphify] warning: ollama returned very few tokens — likely causes: "
            "(1) VRAM pressure: check `nvidia-smi` and reduce chunk size with "
            "--token-budget (e.g. --token-budget 4096) or set "
            "GRAPHIFY_OLLAMA_NUM_CTX to a smaller value; "
            "(2) model too small for JSON instruction following — "
            "try a larger model with --model (e.g. --model qwen2.5-coder:14b).",
            file=sys.stderr,
        )
    return result


def _call_claude(api_key: str, model: str, user_message: str, max_tokens: int = 8192, *, deep_mode: bool = False, images: list[_ImageRef] | None = None) -> dict:
    """Call Anthropic Claude directly (not via OpenAI compat layer)."""
    try:
        import anthropic
    except ImportError as exc:
        raise ImportError(_backend_pkg_hint("anthropic", "anthropic")) from exc

    client = anthropic.Anthropic(
        api_key=api_key,
        base_url=BACKENDS["claude"]["base_url"],
        timeout=_resolve_api_timeout(),
        max_retries=_resolve_max_retries(),
    )
    resp = client.messages.create(
        model=model,
        max_tokens=max_tokens,
        system=_extraction_system(deep=deep_mode),
        messages=[{"role": "user", "content": _anthropic_content(user_message, images or [])}],
    )
    raw_content = resp.content[0].text if resp.content else None
    result = _parse_llm_json(raw_content or "{}")
    result["input_tokens"] = resp.usage.input_tokens if resp.usage else 0
    result["output_tokens"] = resp.usage.output_tokens if resp.usage else 0
    result["model"] = model

View on GitHub (pinned to 7fe58b0b0f)

Solutions

1. Install the extra: `uv tool install "graphifyy[anthropic]" --force` or `pip install anthropic` in the active venv.
2. Confirm with `python -c "import anthropic"` using the same interpreter that runs graphify.
3. Alternatively switch to `backend="claude-cli"` if you have the Claude Code CLI installed and don't want the SDK.

Example fix

# before
pip install graphifyy

# after
pip install "graphifyy[anthropic]"

import importlib.util

if importlib.util.find_spec("anthropic") is None:
    raise SystemExit("graphify's claude backend needs: pip install graphifyy[anthropic]")

try:
    extract_files_direct(files, root, backend="claude")
except ImportError as e:
    if "anthropic" in str(e):
        subprocess.run([sys.executable, "-m", "pip", "install", "anthropic"], check=True)
        extract_files_direct(files, root, backend="claude")
    else:
        raise

When it happens

Trigger: Calling `extract_files_direct(..., backend="claude")` (or `detect_backend()` returning `claude` because ANTHROPIC_API_KEY is set) in an environment without the `anthropic` package installed.

Common situations: Installing graphify without the `[anthropic]` extra but having ANTHROPIC_API_KEY exported, so auto-detection selects `claude`; venv drift after a reinstall; CI images that trim optional dependencies.

Related errors

• the '{pkg}' package is required for this backend but is not
• Azure OpenAI requires the openai package. Run: pip install o
• AWS Bedrock extraction requires boto3. Run: pip install grap
• matplotlib not installed. Run: pip install matplotlib
• neo4j driver not installed. Run: pip install neo4j

AI-assisted analysis of Graphify-Labs/graphify@7fe58b0b0f (2026-08-14). Data as JSON: /api/errors/f01aee6ac197b0ff. Report an issue: GitHub.