feat(parasail): add Parasail as a JSON-configured OpenAI-compatible provider

[Kaihuang724]

Relevant issues

N/A. Adds Parasail (https://parasail.io) as a JSON-configured OpenAI-compatible provider.

Linear ticket

Pre-Submission checklist

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

• I have added meaningful tests
• 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:

Screenshots / Proof of Fix

End-to-end smoke against a live LiteLLM proxy on localhost:4000 hitting the real Parasail API (https://api.parasail.io/v1). The proxy was started with a one-model config:

model_list:
  - model_name: parasail-llama
    litellm_params:
      model: parasail/parasail-llama-33-70b-fp8
      api_key: os.environ/PARASAIL_API_KEY
litellm_settings:
  drop_params: true
general_settings:
  master_key: sk-1234

1. Non-streaming chat completion

curl http://localhost:4000/v1/chat/completions \
  -H "Authorization: Bearer sk-1234" \
  -H "Content-Type: application/json" \
  -d '{"model":"parasail-llama","messages":[{"role":"user","content":"reply with exactly one word: PONG"}],"max_completion_tokens":20}'

{"id":"chatcmpl-bbd11fe9bf31fc4b","created":1780760080,"model":"parasail-llama","object":"chat.completion","choices":[{"finish_reason":"stop","index":0,"message":{"content":"PONG","role":"assistant"}}],"usage":{"completion_tokens":3,"prompt_tokens":43,"total_tokens":46}}

2. Streaming chat completion

curl -N http://localhost:4000/v1/chat/completions \
  -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" \
  -d '{"model":"parasail-llama","messages":[{"role":"user","content":"count from 1 to 3, one number per line"}],"max_completion_tokens":30,"stream":true}'

data: {"choices":[{"index":0,"delta":{"role":"assistant","content":""}}]}
data: {"choices":[{"index":0,"delta":{"content":"1"}}]}
data: {"choices":[{"index":0,"delta":{"content":"\n"}}]}
data: {"choices":[{"index":0,"delta":{"content":"2"}}]}
data: {"choices":[{"index":0,"delta":{"content":"\n"}}]}
data: {"choices":[{"index":0,"delta":{"content":"3"}}]}
data: {"choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}
data: [DONE]

3. Responses API with caller-passed store: true. Parasail returns 400 State storage not supported when store: true reaches their server; LiteLLM forces it to false for Parasail, so the call still succeeds end-to-end.

curl http://localhost:4000/v1/responses \
  -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" \
  -d '{"model":"parasail-llama","input":"reply with exactly one word: PONG","store":true}'

{"id":"resp_YnIXbGVNINBd...","model":"parasail-llama","object":"response","output":[{"id":"msg_b58cffdbb2b0a53a","content":[{"annotations":[],"text":"PONG","type":"output_text"}],"role":"assistant","status":"completed","type":"message"}],"status":"completed","usage":{"input_tokens":44,"output_tokens":3,"total_tokens":47}}

4. Responses API with store omitted. Parasail also rejects requests that omit store (it defaults true server-side); the force-to-false shim covers this too.

curl http://localhost:4000/v1/responses \
  -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" \
  -d '{"model":"parasail-llama","input":"reply with the single word PING"}'

{"id":"resp_ltGX8C5PS2y1...","model":"parasail-llama","object":"response","output":[{"id":"msg_b4a939929002f7f5","content":[{"annotations":[],"text":"PING","type":"output_text"}],"role":"assistant","status":"completed","type":"message"}],"status":"completed","usage":{"input_tokens":43,"output_tokens":2,"total_tokens":45}}

For comparison, the same Responses calls against https://api.parasail.io/v1 directly without the LiteLLM shim return HTTP 400 {"error":{"message":"State storage not supported. Set 'store': false"}} whether the caller passes store: true or omits the field.

Type

New Feature

Changes

Registers parasail in the existing JSON provider loader at litellm/llms/openai_like/providers.json, alongside both /v1/chat/completions and /v1/responses as supported endpoints. The Responses API gateway rejects store: true and rejects requests that omit store (it defaults true server-side), so the JSON loader gains one new special_handling knob, force_store_false. When that flag is set, the dynamically generated Responses config overrides store=False in transform_responses_api_request before delegating to the OpenAI Responses base class. The Parasail entry sets that flag; no other JSON-configured provider opts into it, so existing behavior is unchanged.

Also adds the PARASAIL enum value in LlmProviders, includes parasail in openai_compatible_providers, registers parasail in provider_endpoints_support.json (chat completions + responses), and adds a focused regression test file at tests/test_litellm/llms/parasail/test_parasail.py that covers JSON registry presence, prefix-based provider resolution, chat and Responses URL construction, PARASAIL_API_BASE override behavior, validate_environment header injection, and the force_store_false regression in both the caller-sent-store=true and caller-omitted cases.

User-facing provider documentation will land separately in the litellm-docs repository.

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Greptile Summary

This PR registers Parasail as a JSON-configured OpenAI-compatible provider, following the same pattern used by other providers like neosantara and crusoe. The sole structural novelty is the force_store_false special-handling knob, which intercepts transform_responses_api_request to force store=False before delegating to the parent — needed because Parasail's Responses API rejects requests where store is true or omitted.

• providers.json — adds the parasail entry with base URL, env-var names, supported endpoints, and force_store_false: true.
• dynamic_config.py — extends JSONProviderResponsesConfig with a transform_responses_api_request override that applies force_store_false; the override is generic and will silently no-op for any future provider that does not set the flag.
• Tests — all unit tests, no real network calls; covers registry presence, provider resolution, URL construction, env-var override, header injection, and the force_store_false behaviour for both store=true and omitted cases.

Confidence Score: 5/5

Safe to merge; the change is additive, well-isolated, and follows the existing JSON provider pattern throughout.

All six changed files make minimal, additive changes consistent with how other JSON-configured providers are registered. The only non-trivial logic is the force_store_false override in dynamic_config.py, which is correctly gated behind the provider config flag and tested end-to-end. No existing behaviour is modified.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/llms/openai_like/dynamic_config.py	Adds transform_responses_api_request override to JSONProviderResponsesConfig; implementation correctly gates mutation on provider.special_handling flag and delegates to super(). ResponseInputParam import moved inside the function to avoid circular imports.
litellm/llms/openai_like/providers.json	Adds parasail entry with correct base_url, api_key_env, api_base_env, supported_endpoints, and force_store_false flag; consistent with other JSON-declared providers.
litellm/types/utils.py	Adds PARASAIL = "parasail" to LlmProviders enum; straightforward addition.
litellm/constants.py	Adds "parasail" to openai_compatible_providers list; consistent with other JSON-configured providers already in the list.
provider_endpoints_support.json	Adds parasail endpoint support entry with chat_completions and responses set to true; the documentation URL points to a page that does not yet exist, as acknowledged in the PR description.
tests/test_litellm/llws/parasail/test_parasail.py	Pure unit tests with no real network calls; covers all major behaviours including the force_store_false regression for both caller-set and omitted store values.

_{Reviews (2): Last reviewed commit: "fix(parasail): register in provider_endp..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[Kaihuang724]

@greptileai