docs(guardrails): add during_call mode to Model Armor guardrail docs

[wadhah101]

Summary

The current Model Armor documentation is outdated — it only lists pre_call and post_call modes, but during_call support was added in #15970.

This PR updates the docs to reflect all three supported modes.

Changes

• Added during_call mode to the YAML example, supported values section, and common params
• Fixed post_call description from "input & output" to "output" (it only checks the model response)

Code References

• Default event hooks include all three modes: model_armor.py L65-69
• during_call handler (async_moderation_hook): model_armor.py L469-562

Test plan

• Verify docs render correctly on the docs site

🤖 Generated with Claude Code

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[veria-ai]

Low: Documentation-only change

This PR updates documentation to add during_call mode to the Model Armor guardrail docs. No code changes, no security impact.

---

Status: 0 open
Risk: 1/10

Posted by Veria AI · 2026-04-24T06:46:24.029Z

[greptile-apps]

Greptile Summary

This docs-only PR adds the missing during_call mode to the Model Armor guardrail documentation and corrects the post_call description. Both changes are verified accurate against the implementation in model_armor.py (L65-69 for supported hooks, L469-562 for async_moderation_hook).

Confidence Score: 5/5

Safe to merge — documentation-only change with accurate, verified content.

Single docs file changed; both the addition of during_call and the correction of the post_call description are confirmed correct by reading the source code. No logic, security, or behavioral changes.

No files require special attention.

Important Files Changed

Filename	Overview
docs/my-website/docs/proxy/guardrails/model_armor.md	Adds during_call mode to example config, mode descriptions, and parameter docs; fixes post_call description from "input & output" to "output" — all verified accurate against model_armor.py source.

Sequence Diagram

sequenceDiagram
    participant Client
    participant LiteLLM
    participant ModelArmor
    participant LLM

    Note over LiteLLM,ModelArmor: pre_call mode
    Client->>LiteLLM: Request
    LiteLLM->>ModelArmor: Sanitize input
    ModelArmor-->>LiteLLM: OK / Block
    LiteLLM->>LLM: Forward (if OK)

    Note over LiteLLM,LLM: during_call mode
    Client->>LiteLLM: Request
    LiteLLM->>ModelArmor: Sanitize input (parallel)
    LiteLLM->>LLM: Forward (parallel)
    ModelArmor-->>LiteLLM: OK / Block
    LLM-->>LiteLLM: Response

    Note over LiteLLM,ModelArmor: post_call mode
    Client->>LiteLLM: Request
    LiteLLM->>LLM: Forward
    LLM-->>LiteLLM: Response
    LiteLLM->>ModelArmor: Sanitize output
    ModelArmor-->>LiteLLM: OK / Block
    LiteLLM-->>Client: Response (if OK)

Loading

_{Reviews (2): Last reviewed commit: "docs: add during_call mode to Model Armo..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Congrats! CodSpeed is installed 🎉

🆕 16 new benchmarks were detected.

You will start to see performance impacts in the reports once the benchmarks are run from your default branch.

Detected benchmarks

• test_get_model_info_openai: 81.1 µs
• test_token_counter_raw_text: 183.8 µs
• test_get_llm_provider_with_prefix: 136.8 µs
• test_token_counter_simple_message: 235 µs
• test_token_counter_with_tools: 414.1 µs
• test_get_llm_provider_openai: 138.3 µs
• test_get_model_cost_key_case_insensitive: 81.9 µs
• test_get_llm_provider_anthropic: 142.4 µs
• test_get_model_info_with_provider: 79.7 µs
• test_get_model_info_anthropic: 80.2 µs
• test_token_counter_multi_turn: 567.6 µs
• test_get_llm_provider_azure: 144.2 µs
• test_token_counter_long_content: 1.7 ms
• test_get_model_cost_key_exact_match: 79.7 µs
• test_cost_per_token_openai: 540 µs
• test_cost_per_token_anthropic: 541.1 µs

---