cruxial 0.4.0

The reliability layer for LLM tool calls. Intercepts every call, validates arguments against your schema, auto-repairs hallucinated args before execution.

Cruxial

The reliability layer for LLM tool calls.

Your agent said it sent the email. It didn't.

Cruxial intercepts every LLM tool call before it executes. It validates the arguments against your schema and auto-repairs hallucinated args with a structured retry. Drop-in for OpenAI and Anthropic. Overhead is under 1ms p99, because validation runs locally with no extra network hop. It fails open by default: if Cruxial itself errors, your tool still runs.

pip install cruxial

Then watch it catch every failure category live — offline, no API key:

cruxial demo

30-second demo

import json
from cruxial import guard
from openai import OpenAI

client = OpenAI()

# Standard OpenAI tool definitions
schemas = {
    "send_email": {
        "type": "object",
        "properties": {
            "to": {"type": "string", "format": "email"},
            "subject": {"type": "string", "maxLength": 200},
            "body": {"type": "string"},
        },
        "required": ["to", "subject", "body"],
    }
}

# Your actual executors
def send_email(to, subject, body):
    return mailer.send(to=to, subject=subject, body=body)

executors = {"send_email": send_email}

# Wrap once
cruxial = guard(schemas=schemas, executors=executors)

# In your agent loop:
for tool_call in llm_response.tool_calls:
    args = json.loads(tool_call.arguments)   # OpenAI returns arguments as a JSON string
    result = cruxial.execute(tool_call.name, args)

    if not result.ok:
        # result.failure.category says why (e.g. "type_mismatch"); for an executor
        # error the raw exception is on result.error.
        result.raise_on_failure()        # raises the right typed error either way

    use(result.value)

Or: one call does the whole turn

guard() gives you full control. If you write the usual raw-SDK loop, cruxial.run() does the entire tool step in one call: it calls the model, validates every tool call, executes the valid ones, auto-repairs the bad ones in one round-trip, then logs and appends the results. Works with OpenAI, Azure, Anthropic, and LiteLLM. You keep your loop:

import cruxial

result = cruxial.run(
    client,                 # your OpenAI / AzureOpenAI / Anthropic client, or litellm.completion
    model="gpt-4o",
    messages=messages,
    tools=tools,            # the same tool defs you already pass the LLM
    executors=executors,    # {tool_name: your_function}
)

while not result.finished:  # your loop stays yours — one model call per turn
    result = cruxial.run(client, model="gpt-4o", messages=result.messages,
                         tools=tools, executors=executors)

print(result.text)          # the model's final answer

It reuses your configured client (Azure endpoint, base_url, timeouts all preserved), derives schemas from tools, and fails open. Deliberately one turn, not a framework — no streaming, no multi-turn ownership, you decide when to stop. Need to own execution? Drop to guard().check() / .execute().

Async? Use await cruxial.arun(...) with an AsyncOpenAI / async client, and await guard().aexecute(...) for async tool executors — same contract, everything awaited.

What it catches

Eight failure categories. Every interception is logged with the failure category — never the raw argument values.

Category	What it catches
missing_required	Required field not in args
type_mismatch	Wrong type (int instead of str, etc.)
enum_violation	Value not in allowed enum
format_violation	Bad email / uri / date format, or a pattern mismatch
constraint_violation	maxLength / minimum / multipleOf / etc.
extra_field	Model invented a field that doesn't exist
unknown_tool	Tool name not in registry
tool_bypass	Model claimed it did something but emitted no call — "your agent said it sent the email. It didn't."

The first seven are schema-derivable. tool_bypass is the one validators structurally can't catch — there's no call to validate. See below.

tool_bypass — catch the action your agent claimed but never took

A model says "I've sent the email" and emits no send_email call. No call ever goes out — so no error, no log, nothing to grep for. The silent failure. cruxial.run() catches it:

result = cruxial.run(client, model="gpt-4o", messages=messages,
                     tools=tools, executors=executors)   # bypass check is on by default

if result.bypass:        # the model claimed an action and, when re-prompted, confirmed it
    print("caught a bypass:", result.bypass.tool)

How it works (zero cost on normal turns): a final text turn is flagged only when it claims a completed action ("sent", not "send"), attributed to the assistant (not "you" / "the scheduler" / "automatically"), for a side-effecting tool that was never called. A flagged turn gets one neutral re-prompt — the model either re-emits the call (corrected + executed) or declines (we do nothing). We only ever act on a model-confirmed re-emission, so we never fabricate an action.

Benchmarked on a 132-scenario adversarial set (BENCHMARKS.md): 0 false actions, acted-on precision 100% (sonnet-4-6 and gpt-4o), 100% correction recall on that set. It's precision-first — it fires on completion-form verbs, so terse claims ("Done.", "Email's out.") are a documented recall gap: a high-quality net, not a complete guarantee. bypass="off" disables it; bypass="strict" is a 2-call variant.

Auto-repair

cruxial.run() already auto-repairs for you. If you use guard() directly, call the adapter helper with the failure context:

from cruxial.adapters.openai import auto_repair

cruxial = guard(schemas=schemas, executors=executors)
result = cruxial.execute(name, args)

if not result.ok:
    # 1-attempt structured retry — feeds the failure back to the model
    new_args = auto_repair(
        client,
        model=model,
        messages=messages,            # conversation incl. the assistant tool-call turn
        tools=tools,                  # the same tool defs you sent the model
        failure=result.failure,
        failed_args=args,
        repair_prompt=cruxial.build_repair_prompt(result.failure, args),
    )
    result = cruxial.execute_repaired(name, new_args)

Roughly 90% of intercepted calls are fixed in a single repair round-trip on the pooled live-MCP benchmark (87–94% per run on current code). Every number is sourced in BENCHMARKS.md.

See your interception rate

Every interception is written to a local SQLite file (no data leaves your machine). To see your real rate:

cruxial stats

Stats are project-local automatically — inside a project (a dir with .git/, pyproject.toml, setup.py, or .cruxial/) the DB lives at ./.cruxial/telemetry.sqlite, so each app stays separate; elsewhere it falls back to ~/.cruxial/. Override with CRUXIAL_DB_PATH; cruxial diagnostic shows the active path.

Output:

cruxial · last 24h
─────────────────────────────────────────
  total calls           1,247
  intercepted             184  (14.8%)
  auto-repaired           167  (90.8% of intercepted)
  passed through        1,063

top failing tools                    rate
  send_email                        23.1%
  create_calendar_event             18.4%
  search_web                         9.2%

top failure categories
  type_mismatch                       62
  missing_required                    44
  enum_violation                      38
  format_violation                    24
  constraint_violation                16

cruxial stats also shows the registry independently of traffic — handy for the "is it even on?" check after install:

  registry              6 registered  ·  3 fired  ·  1 intercepted

Traffic but 0 interceptions usually means a well-behaved model on a simple schema — cruxial.testing.violation_payloads(schema) fires a synthetic violation per category to verify end-to-end.

How it works

sequenceDiagram
    participant M as LLM model
    participant C as Cruxial guard
    participant T as Your tool / executor
    participant DB as Local SQLite telemetry

    M->>C: tool call (name, args)
    C->>C: validate args vs JSON Schema
    alt args valid
        C->>DB: record PASSED (hashes only)
        C->>T: run tool
        T-->>M: result
    else args invalid
        C->>DB: record INTERCEPTED + failure category
        opt auto-repair enabled
            C-->>M: repair prompt (1-shot retry)
            M->>C: corrected tool call
        end
        C-->>M: typed failure (caller decides what to do)
    end

Cruxial wraps the tool registry, not the LLM client. No monkey-patching, no proxies, no framework lock-in.

Schema source — if your LLM sees a trimmed schema

If you keep two views of a tool schema — a canonical one for execution and a trimmed one sent to the LLM — register the trimmed one. The LLM can only satisfy the schema it was shown; validating against canonical fields it never saw turns missing_required / extra_field into false positives.

# Right
cruxial = guard(schemas=tool_definitions_sent_to_llm)

# Must use the canonical schema? Opt in — validation is unchanged, but it warns at
# construction and tags every telemetry row (visible in `cruxial stats`) to filter later:
cruxial = guard(schemas=canonical_definitions,
                config=GuardConfig(schema_origin="canonical"))

Privacy

By default Cruxial stores: tool name, schema fingerprint, failure category, timing. Never the argument values themselves. Hashes only.

The interceptor runs in your process. Your data never leaves your infrastructure unless you opt into Cruxial Cloud (coming soon).

Define tools as Pydantic models

Already model your tools with Pydantic? Skip the hand-written JSON Schema — cruxial extracts it for you. Optional: pip install cruxial[pydantic] (v2).

from pydantic import BaseModel, Field
from cruxial.adapters.pydantic import guard_models, tool_schema

class SendEmail(BaseModel):
    """Send an email to a recipient."""
    to: str = Field(pattern=r"^[^@\s]+@[^@\s]+\.[^@\s]+$")
    subject: str = Field(min_length=1, max_length=200)
    body: str

# One source of truth → both the runtime guard and the LLM tool definition
cruxial = guard_models({"send_email": SendEmail}, executors={"send_email": send_email})
tools   = [tool_schema(SendEmail, name="send_email")]                       # OpenAI format
# tools = [tool_schema(SendEmail, name="send_email", provider="anthropic")]  # Anthropic

Works with run() too — run() derives the schema from tools, so the whole managed turn (validate → execute → auto-repair) runs against your model:

from cruxial import run
run(client, model="gpt-4o", messages=msgs,
    tools=[tool_schema(SendEmail, name="send_email")],
    executors={"send_email": send_email})

Nested models and enums (Pydantic's $defs/$ref) validate end to end. The adapter is lazy and optional — cruxial's core never requires Pydantic.

What ships today (v0.4)

• ✅ Python SDK
• ✅ OpenAI + Azure OpenAI + Anthropic + LiteLLM (auto via normalization)
• ✅ JSON Schema validation
• ✅ 7 schema-validation categories (+ tool_bypass below = 8 total)
• ✅ 1-attempt auto-repair
• ✅ Local SQLite + stdout telemetry
• ✅ cruxial stats CLI
• ✅ Fail-open by default
• ✅ cruxial.run() — one managed turn (OpenAI / Azure / Anthropic / LiteLLM)
• ✅ tool_bypass detection — the claimed-but-never-called catch
• ✅ Pydantic adapter — define tools as Pydantic models (cruxial[pydantic])

Coming:

• TypeScript SDK
• LangChain, LlamaIndex, AutoGen adapters
• Hosted dashboard with cross-customer schema drift alerts
• Zod (TypeScript) validators

Questions or feedback?

Open a GitHub issue, or reach the maintainers on Discord. We'd especially like to hear from anyone running Cruxial in a real agent — early design-partner feedback shapes the roadmap.

License

MIT. The SDK runs entirely in your process. The hosted dashboard (Cruxial Cloud) will be a separate paid product. The interceptor itself stays MIT forever.