Compatibility

LM0 prioritizes practical SDK compatibility over implementing every optional field of the OpenAI and Anthropic APIs.

Client surfaces#

Surface	Path	Typical clients
OpenAI Chat Completions	POST /v1/chat/completions	OpenAI SDKs, LiteLLM, many agent frameworks
OpenAI Responses	POST /v1/responses	Newer OpenAI clients (subset)
Anthropic Messages	POST /v1/messages	Anthropic SDKs, Claude-compatible tools
Models	GET /v1/models	Both ecosystems (shape via headers)

Auth compatibility#

Style	Header
OpenAI	Authorization: Bearer <key>
Anthropic	x-api-key: <key>

Provider compatibility#

Upstream adapters speak:

• OpenAI-compatible HTTP APIs
• Anthropic-compatible HTTP APIs

Cross translation is first-class: OpenAI clients can target Anthropic providers and Anthropic clients can target OpenAI providers via the canonical middle layer.

Supported feature areas#

Feature	Support
Non-streaming text	Yes
Streaming (SSE)	Yes (surface-specific)
Vision / image input	When upstream supports it
Tool / function calling	Translated where possible
Reasoning / thinking	Visible translation where upstream exposes it
Model listing	Yes (dual shape)
Token counting	Anthropic route; estimates for some targets

Known limitations#

The gateway intentionally does not implement every OpenAI or Anthropic field. High-level gaps include:

• Incomplete coverage of niche sampling / modality parameters
• Responses API is a stateless subset, not full stateful Responses
• Some tool and reasoning edge cases depend on upstream fidelity
• Provider-specific extensions may be dropped during translation

When in doubt, start with simple chat/messages payloads, then enable tools and streaming.

Billing & enforcement (DB mode)#

Concern	Behavior
Credits / package tokens	Reserved and recorded around generation
RPM	Per account tier
Ban	Hard block
Maintenance pause	Blocks generation endpoints only

Related#

• API overview
• Architecture