feat(providers): add Bitdeer AI as a JSON-configured provider

[ruijietey]

TLDR

Problem this solves:

• Bitdeer AI's OpenAI-compatible inference API has no named provider in LiteLLM
• Callers must manually pass a custom api_base/api_key today

How it solves it:

• Registers bitdeer-ai in the JSON-configured provider system (providers.json)
• Adds pricing for 3 models to model_prices_and_context_window.json

User Flow

Before: a developer wants to call a Bitdeer-hosted model through LiteLLM

1. They call litellm.completion(model="moonshotai/Kimi-K3", api_base="https://api-inference.bitdeer.ai/v1", api_key=os.environ["BITDEER_API_KEY"])
2. This works, but there is no named "bitdeer-ai" provider, no cost tracking, and no auto-detection from api_base alone

After: the same developer gets a named provider with cost tracking

1. They call litellm.completion(model="bitdeer-ai/moonshotai/Kimi-K3", api_key=os.environ["BITDEER_API_KEY"]) with no api_base needed
2. The request resolves to https://api-inference.bitdeer.ai/v1/chat/completions and the response includes cost calculated from model_prices_and_context_window.json
3. Passing only api_base="https://api-inference.bitdeer.ai/v1" with a bare model name also auto-resolves to custom_llm_provider="bitdeer-ai"

Relevant issues

Linear ticket

Pre-Submission checklist

• I have added meaningful tests
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• My PR passes all CI/CD checks (e.g., lint, format, unit tests) — ran locally (pytest, schema validation against model_prices_and_context_window.schema.json, ruff check scoped to touched litellm/ files); have not run full CI
• I have received a Greptile Confidence Score of at least 4/5 — pending, will request after this opens

Screenshots / Proof of Fix

No live Bitdeer API key was available while preparing this PR, so there is no real end-to-end curl proof against the live endpoint. What is verified locally:

$ python3 -m pytest tests/test_litellm/llms/openai_like/test_bitdeer_provider.py -v
8 passed

$ python3 -c "import json, jsonschema; jsonschema.validate(json.load(open('model_prices_and_context_window.json')), json.load(open('model_prices_and_context_window.schema.json'))); print('schema valid')"
schema valid

Happy to run a real curl against a live Bitdeer key if a maintainer or the Bitdeer team can supply one for verification before merge.

Type

🆕 New Feature

Caveats (if any)

• No live API key was available to produce real end-to-end proof, see above
• Only 3 of Bitdeer's models are registered here; more can follow in a separate PR

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

[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.}

[greptile-apps]

Greptile Summary

This PR registers Bitdeer AI as a JSON-configured OpenAI-compatible provider and adds pricing and capability metadata for three models.

• Adds provider identification, URL autodetection, environment-key lookup, and OpenAI-compatible dispatch registration.
• Adds model pricing/context metadata and focused provider-resolution tests.
• The URL-autodetection credential precedence needs correction, and one pricing key is duplicated.

Confidence Score: 4/5

The provider wiring is mostly complete, but the explicit credential precedence defect should be fixed before merging.

Bitdeer requests using api_base autodetection can authenticate with BITDEER_API_KEY instead of the caller’s explicit key when both are present; the pricing registry also contains a non-blocking duplicate key.

Files Needing Attention: litellm/litellm_core_utils/get_llm_provider_logic.py; model_prices_and_context_window.json

Important Files Changed

Filename	Overview
litellm/litellm_core_utils/get_llm_provider_logic.py	Adds Bitdeer endpoint autodetection, but the branch can replace an explicit API key with BITDEER_API_KEY.
litellm/llms/openai_like/providers.json	Correctly registers Bitdeer's base URL, credential environment variable, and token-parameter mapping.
model_prices_and_context_window.json	Adds three Bitdeer pricing records, but DeepSeek-V4-Flash is declared twice under the same key.
litellm/types/utils.py	Consistently adds bitdeer-ai to the provider enum.
litellm/constants.py	Consistently adds Bitdeer to endpoint and OpenAI-compatible provider registries.
tests/test_litellm/llms/openai_like/test_bitdeer_provider.py	Covers registration, routing, URL detection, and pricing, but does not exercise URL autodetection with competing explicit and environment keys.

_{Reviews (1): Last reviewed commit: "Trim Bitdeer models to 3 and rename slug..." | Re-trigger Greptile}

[greptile-apps]

Explicit API key gets overwritten

If BITDEER_API_KEY is set, URL autodetection replaces the caller's explicit api_key with that environment value, causing authentication to use the wrong Bitdeer account or fail despite a valid explicit credential.

Suggested change

	elif endpoint == "https://api-inference.bitdeer.ai/v1":
	custom_llm_provider = "bitdeer-ai"
	dynamic_api_key = get_secret_str("BITDEER_API_KEY")
	elif endpoint == "https://api-inference.bitdeer.ai/v1":
	custom_llm_provider = "bitdeer-ai"
	dynamic_api_key = api_key or get_secret_str("BITDEER_API_KEY")

Knowledge Base Used: LLM Provider Adapters

[greptile-apps]

Duplicate model pricing key

bitdeer-ai/deepseek-ai/DeepSeek-V4-Flash is declared twice, so standard JSON loading silently discards the first capability record and existing schema and pricing tests cannot detect edits made to it.

Suggested change

	"bitdeer-ai/deepseek-ai/DeepSeek-V4-Flash": {
	"supports_prompt_caching": true,
	"source": "https://www.bitdeer.ai/en/pricing/ai-models"
	},

Knowledge Base Used: Cost Tracking and Budget Enforcement

[veria-ai]

Medium: API key exfiltration through an overridden base URL

A caller can submit model: "bitdeer-ai/..." with api_base pointing to a server they control and omit api_key. The JSON-provider resolver preserves that api_base but falls back to BITDEER_API_KEY, so the outbound request exposes the server's credential in its Authorization header. Reject untrusted base-URL overrides for this provider, or require an explicit request API key whenever the selected base URL differs from the configured Bitdeer URL.

[veria-ai]

PR overview

This pull request adds Bitdeer AI as an OpenAI-compatible provider configured through the JSON provider registry.

One security issue remains open: callers can override the Bitdeer API base URL while the provider still falls back to the server’s configured Bitdeer API key. This can expose that credential to an attacker-controlled endpoint, so the provider should restrict base-URL overrides or require an explicit key for nonstandard URLs.

Open issues (1)

• Medium: API key exfiltration through an overridden base URL — litellm/llms/openai_like/providers.json:189

Fixed/addressed: 0 · PR risk: 6/10

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing ruijietey:litellm_add_bitdeer_provider (e4ab197) with litellm_internal_staging (d86336a)}