[ghost]

Relevant issues

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/test_litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Apr 3, 2026 9:03pm

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 8 committers have signed the CLA.

✅ madhu19991
❌ devarakondasrikanth
❌ ishaan-berri
❌ michelligabriele
❌ redhelix
❌ Sameerlite
❌ talhaanwarch
❌ github-actions[bot]

---

ishaan-berri seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[codspeed]

Merging this PR will not alter performance

✅ 16 untouched benchmarks

---

_{Comparing litellm_ishaan_april1 (1719bfa) with main (fc885af)}

[greptile-apps]

Greptile Summary

This PR delivers a multi-area set of improvements across the proxy, LLM integrations, and UI:

• Environment-scoped prompt management: Adds environment and created_by columns to LiteLLM_PromptTable (migration + all 3 schema.prisma copies kept in sync). All prompt CRUD endpoints now accept an ?environment= filter. The unique constraint on (prompt_id, version) is widened to (prompt_id, version, environment) so the same version number can exist independently in each environment.

• Amazon Nova Canvas image editing on Bedrock: New BedrockAmazonNovaCanvasImageEditConfig transformation class routed via a supports_nova_canvas_image_edit flag in model_prices_and_context_window.json, following the codebase's config-driven model capability pattern.

• user_custom_key_update hook: Adds a new config option general_settings.custom_key_update analogous to custom_key_generate. Wired into single-key /key/update and /key/bulk_update flows.

• _enforce_upperbound_key_params refactor: Extracts inline upperbound enforcement into a reusable helper; applies to key updates with fill_defaults=False.

• Bedrock Anthropic streaming cache-token fix: _promote_message_stop_usage buffers message_delta and merges cache fields from message_stop, fixing under-reported cache usage for clients that only read message_delta.

• Firecrawl self-hosted (v1) support: Response parser now handles flat-list data arrays from self-hosted Firecrawl alongside v2 cloud dict format.

• Failed-request duration fix: proxy_track_cost_callback reads actual start_time from the logging object for failed requests.

Key observations: all three schema.prisma copies are updated consistently; migration SQL is correct and backward-compatible; tests are unit-only with no real network calls.

Confidence Score: 5/5

Safe to merge; all findings are P2 style/quality suggestions with no blocking correctness issues.

All three schema.prisma files are in sync, the migration is backward-compatible, tests are fully mocked, and the key-update hook integration is clean with no double-invocation. Remaining findings are P2: a silent exception swallow in _get_prompt_template, an extra unconditional DB call per GET /prompts/{id}, and an inline import litellm that violates the CLAUDE.md style guide.

litellm/proxy/prompts/prompt_endpoints.py — extra DB call and silent exception swallow; litellm/llms/bedrock/image_edit/amazon_nova_canvas_image_edit_transformation.py — inline import style

Important Files Changed

Filename	Overview
litellm/proxy/prompts/prompt_endpoints.py	Adds environment filtering to list/get/version endpoints and switches get_prompt_versions to query the DB directly; introduces an extra DB call for all_environments in get_prompt_info; _get_prompt_template silently swallows all exceptions
litellm/proxy/management_endpoints/key_management_endpoints.py	Extracts _enforce_upperbound_key_params helper and integrates user_custom_key_update hook; refactoring is logically equivalent to original; hook called once per code path
litellm/llms/bedrock/image_edit/amazon_nova_canvas_image_edit_transformation.py	New file adding Nova Canvas image-edit support; uses model_prices JSON flag for routing; inline import of litellm inside _supports_nova_canvas_image_edit_from_model_cost violates CLAUDE.md style guide
litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py	Adds _promote_message_stop_usage async generator to fix cache-token accounting for Bedrock Anthropic streaming; logic correctly buffers message_delta and merges message_stop usage fields
litellm/llms/firecrawl/search/transformation.py	Adds branch to handle Firecrawl v1 self-hosted flat-list response alongside existing v2 cloud dict format; clean implementation
litellm-proxy-extras/litellm_proxy_extras/migrations/20260331000000_add_prompt_environment_and_created_by/migration.sql	Adds environment (NOT NULL DEFAULT 'development') and created_by columns, drops old unique index, creates new composite unique and index; all three schema.prisma files are kept in sync
litellm/proxy/hooks/proxy_track_cost_callback.py	Fixes failed request duration tracking by reading actual start_time from the logging object instead of always using datetime.now()
litellm/proxy/proxy_server.py	Adds user_custom_key_update global variable, wires it from general_settings.custom_key_update, and resets it in cleanup/shutdown handlers
litellm/types/prompts/init_prompts.py	Adds environment and created_by fields to PromptSpec and PromptInfoResponse; adds environments list field to support multi-environment prompt display
tests/test_litellm/llms/bedrock/image_edit/test_amazon_nova_canvas_image_edit.py	New unit tests for Nova Canvas image-edit; correctly uses monkeypatch/mocks, no real network calls

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    subgraph Prompt_Endpoints["Prompt Endpoints (environment-scoped)"]
        A[GET /prompts] -->|environment?| B{filter in-memory or DB}
        C[GET /prompts/versions/:id] -->|environment?| D{DB query or in-memory fallback}
        E[GET /prompts/:id] -->|environment?| F{DB query with environment filter}
        F --> G[extra DB call: all_environments]
        POST_P[POST /prompts] --> H[get_next_version per environment]
        PUT_P[PUT /prompts/:id] --> H
    end
    subgraph Key_Update["Key Update Hook"]
        K1[PUT /key/update] --> K2[_validate_update_key_data]
        K2 --> K3[user_custom_key_update hook]
        K3 --> K4[_enforce_upperbound_key_params fill_defaults=False]
        K5[POST /key/bulk_update] --> K6[_process_single_key_update per key]
        K6 --> K3
        K6 --> K4
    end
    subgraph Bedrock_Nova["Bedrock Nova Canvas Image Edit"]
        N1[image_edit call] --> N2[_is_nova_canvas_image_edit_model reads model_cost JSON flag]
        N2 -->|yes| N3[_nova_canvas_task_body]
        N3 -->|mask?| N4[INPAINTING]
        N3 -->|no mask| N5[IMAGE_VARIATION]
        N3 -->|explicit| N6[OUTPAINTING / BACKGROUND_REMOVAL]
    end
    subgraph Streaming_Fix["Bedrock Anthropic Streaming Fix"]
        S1[completion_stream] --> S2[_promote_message_stop_usage]
        S2 -->|buffer message_delta| S3[message_stop arrives]
        S3 -->|merge cache usage fields| S4[yield patched message_delta]
        S4 --> S5[yield message_stop]
    end

Loading

_{Reviews (2): Last reviewed commit: "fix(lint): extract helpers to fix PLR091..." | Re-trigger Greptile}

[greptile-apps]

Non-coroutine hook raises ValueError, which is caught and repackaged as "Authentication Error"

Both _process_single_key_update (line ~1757) and update_key_fn (line ~2164) raise ValueError("user_custom_key_update must be a coroutine") when the configured hook is a plain (synchronous) function. In update_key_fn, this ValueError is swallowed by the outer except Exception as e: handler and surfaced to the caller as:

{"error": "Authentication Error, user_custom_key_update must be a coroutine"}

A configuration error deserves a more actionable message. Consider raising HTTPException(status_code=500, detail="...") (or a dedicated ProxyException) instead of ValueError, so the existing error handler forwards the correct status code and a clear explanation.