feat(provider): Add Neosantara provider as OpenAI Compatible

[ErRickow]

Relevant issues

This is re open pull request and fix from PR: #20641

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

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

Ran LiteLLM proxy locally and verified Neosantara end-to-end.

Chat completions:

  curl -sS http://127.0.0.1:4000/v1/chat/completions \
    -H 'Authorization: Bearer sk-test' \
    -H 'Content-Type: application/json' \
    -d '{"model":"neosantara-chat","messages":
    [{"role":"user","content":"Reply exactly: litellm-neosantara-opus-chat-
    ok"}],"max_tokens":20}'

Returned: litellm-neosantara-opus-chat-ok

Responses API:

  curl -sS http://127.0.0.1:4000/v1/responses \
    -H 'Authorization: Bearer sk-test' \
    -H 'Content-Type: application/json' \
    -d '{"model":"neosantara-responses","input":"Reply exactly: litellm-
    neosantara-opus-responses-ok","max_output_tokens":40}'

Returned: status: completed, text: litellm-neosantara-opus-responses-ok

NOTE: some models in free tier may can't be used, use gemini-3-flash-preview, garda-code, grok-4.1-fast-non-reasoning for testing :)

Type

🆕 New Feature
🧹 Refactoring
📖 Documentation
✅ Test

Changes

docs/my-website/docs/providers/neosantara.md
litellm/llms/openai_like/providers.json
litellm/types/utils.py
provider_endpoints_support.json
tests/test_litellm/llms/neosantara/test_neosantara.py

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks

---

_{Comparing neosantara-xyz:litellm_neosantara_provider (d8b4b80) with main (5be0797)}

[greptile-apps]

Greptile Summary

This PR integrates Neosantara as an OpenAI-compatible provider by following the JSON-registry pattern already used by providers such as crusoe and chutes. No custom HTTP handler is introduced; all routing reuses the existing openai_like infrastructure.

• Adds a neosantara entry to providers.json (base URL, env-var names, max_completion_tokens→max_tokens mapping, and supported endpoints) and a matching NEOSANTARA enum value in LlmProviders.
• Registers the provider's endpoint capabilities in provider_endpoints_support.json (chat completions + responses API enabled).
• Adds a mock-only test suite covering registry lookup, dynamic config, provider-prefix routing, URL construction, parameter mapping, and responses-API config.

Confidence Score: 5/5

Safe to merge; changes are additive and isolated to the JSON provider registry, the LlmProviders enum, the endpoint-support manifest, and a new mock test file — no existing behavior is altered.

The implementation uses the same data-driven registry pattern as other recently-added OpenAI-compatible providers, introduces no custom HTTP logic, and the tests are mock-only. The only gaps are a missing init.py in the test directory and the absence of a default-base-URL test case, neither of which affects runtime correctness.

tests/test_litellm/llms/neosantara/test_neosantara.py — missing init.py and a default-fallback test case that peer providers include.

Important Files Changed

Filename	Overview
litellm/llms/openai_like/providers.json	Adds the neosantara provider entry following the established JSON pattern; includes base URL, env var names, max_completion_tokens→max_tokens mapping, and supported_endpoints.
litellm/types/utils.py	Adds NEOSANTARA = "neosantara" to the LlmProviders enum between CHUTES and XIAOMI_MIMO; minimal, correct change.
provider_endpoints_support.json	Adds neosantara endpoint support entry; the docs URL references a page that may not be live until the docs-repo PR is merged.
tests/test_litellm/llms/neosantara/test_neosantara.py	Mock-only test suite covering registry, env-var config, provider detection, URL generation, param mapping, and responses API; missing init.py and a default-base-URL test that peer providers include.

_{Reviews (2): Last reviewed commit: "Address Neosantara provider review feedb..." | Re-trigger Greptile}

[ErRickow]

@mateo-berri hi, reopen and pull to main. Let me know if there's any issues or mybe need more information about neosantara or about the integration 👍🏻

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ad947e101a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Use a Neosantara-listed Responses model

When users copy this Responses example, it sends claude-opus-4-6, but Neosantara's own Models Overview does not list that model ID; the available Anthropic IDs there are in the claude-4.5-* family. In that scenario the new provider docs guide users to an immediate unknown-model/provider rejection even though their LiteLLM configuration is otherwise correct; the proxy sample below repeats the same ID.

Useful? React with 👍 / 👎.

[ErRickow]

it's fine, the actual model is live on api.neosantara.xyz/v1/models, the docs got slow update about the model ID

[chatgpt-codex-connector]

Add Neosantara to the packaged endpoint support file

This only adds Neosantara to the root support JSON, but the proxy endpoint that serves supported endpoints reads litellm/provider_endpoints_support_backup.json via public_endpoints.py:146; I checked that packaged file with rg and it still has no neosantara entry. As a result, /public/supported_endpoints will continue omitting this new provider even though this file says chat and responses are supported.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Map max_completion_tokens to max_tokens

For callers that use LiteLLM's OpenAI chat parameter max_completion_tokens, this JSON config inherits the default OpenAI GPT supported-params list and forwards max_completion_tokens unchanged because no param_mappings entry is defined. Neosantara's chat-completions documentation lists max_tokens as the generation limit parameter, so those otherwise valid LiteLLM calls can be rejected by the provider; add the same max_completion_tokens → max_tokens mapping used by neighboring JSON providers.

Useful? React with 👍 / 👎.

[Sameerlite]

@ErRickow Can you get the score to 5/5?

[ErRickow]

@Sameerlite yup, wait a minute

[ErRickow]

@greptileai re review

[ErRickow]

@Sameerlite hi, the score already 5/5. If you need any details or need some changes again, feel free to discuss

[Sameerlite]

@ErRickow can you resolve conflicts? Thanks!

[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.
2 out of 3 committers have signed the CLA.

✅ yuneng-berri
✅ ErRickow
❌ shin-berri
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[ErRickow]

@Sameerlite hi, already resolved the conflict, you can check now