- Category
- AI Tools
- Rank
- No. 1080Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- supermemoryai
- GitHub
- 328 stars
- Date
About
Universal LLM input-format adapter with built-in observability and error handling — swap models without rewriting prompts.
What it does
LLM Bridge normalizes request bodies and SSE event streams into an intermediate representation, then renders them for OpenAI Chat, OpenAI Responses, Anthropic, or Google. It also carries multimodal content, tool calls, reasoning settings, structured output, and provider errors across those translations. An optional request handler can modify normalized requests before forwarding them.
Why it's ranked here
The scope is unusually complete for a small TypeScript library. Request conversion, streaming, tool lifecycles, multimodal content, structured output, error normalization, and telemetry share one model. The package also has no declared runtime dependencies. The main reservation is precision: token and cost telemetry includes explicit approximations, while model prices depend on a remote dataset.
What's good
The intermediate model creates one place to inspect or alter requests before converting them back. Provider-specific fields can be retained for reconstruction. Streaming has parsers and emitters for every listed format, including OpenAI Responses. Error records preserve provider context and raw input. The empty runtime dependency list keeps adoption friction low.
Tradeoffs
Token counts use character ratios and fixed estimates for images, audio, video, documents, and tool calls. Unknown models receive zero costs and capabilities. Price lookup contacts an external GitHub-hosted dataset and falls back to defaults on failure. The forwarding handler removes the content-type header from the caller-provided header object. Google and OpenAI Responses errors do not receive equally complete parsing coverage.
How to use it well
Use it inside a TypeScript proxy, gateway, or request-processing layer that must inspect prompts and support several provider wire formats. Keep the normalized representation as the boundary for transforms, then translate only at ingress and egress. Treat token and cost data as operational estimates. It does not replace provider credentials, model selection policy, a hosted gateway, or a command-line client.
Technical notes+
src/index.ts re-exports helpers, handlers, errors, models, types, tools, and streaming surfaces. src/models/index.ts dispatches four provider formats through a typed universal body. src/handler.ts detects a provider, normalizes and edits the request, performs a fetch, computes observability data, and translates SSE through provider-specific parser and emitter selection. src/models/helpers.ts uses approximate token accounting and fetches model pricing from an AgentOps GitHub dataset with a 24-hour in-memory cache. src/errors/parser.ts normalizes OpenAI, Anthropic, and Google errors. package.json declares public npm packaging, CommonJS and ES module entry points, TypeScript declarations, no runtime dependencies, and provider SDKs only as development dependencies.
Observed
- License
- MIT
- Primary language
- TypeScript
- Install surface
- Public npm package installed with npm install llm-bridge
- Interface
- TypeScript library with request translation, SSE streaming, error, helper, and observability exports
- Provider formats
- OpenAI Chat Completions, OpenAI Responses, Anthropic Claude, and Google Gemini
- Module packaging
- CommonJS and ES module entry points with TypeScript declarations
- Runtime dependencies
- The package declares an empty dependencies object
- Published contents
- The package manifest limits published files to dist
Read from README.md, package.json, src/index.ts, src/handler.ts, src/tools/index.ts, src/types/index.ts, src/errors/index.ts, src/models/index.ts, src/errors/types.ts, src/errors/utils.ts, src/helpers/index.ts, src/errors/parser.ts, src/helpers/utils.ts, src/models/helpers.ts, src/streaming/index.ts.
What it can do
Adapt prompts between different LLM input formats
Prompt in one LLM format → Prompt converted to target LLM format
Switch between LLM models without rewriting prompts
Existing prompt and target model specification → Model response using adapted prompt format
Monitor LLM API calls and responses
LLM requests and responses → Observability data and metrics
Handle LLM API errors automatically
Failed LLM API requests → Error recovery responses or fallback handling
Standardize prompts across multiple LLM providers
Raw prompt text → Standardized prompt format compatible with multiple models
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.