Twigg

Applications that call large language models can hand the job of remembering a conversation to Twigg, a hosted context layer. Normally the whole conversation is rebuilt and sent again with each request. With Twigg, a developer opens a chat one time and from then on supplies only the next event.

What the service does with the conversation

Twigg stores the conversation outside the model providers and fits it to whichever model is being called. When a conversation runs long, it compacts or truncates it, and then it routes the call. Because the history lives in Twigg, the model can be swapped mid-conversation.

Providers and management

Calls can be sent on to six destinations named in its documentation: Anthropic, OpenAI, Google, xAI, Fireworks and OpenRouter. A dashboard is where tool schemas, system prompts and context windows are set, and it also shows the cost of each request.

API and documentation

The API is a REST service authenticated with a bearer key, and an OpenAPI specification is published. The whole documentation set is also available as plain markdown so that coding agents can read it.

Pricing approach

The prices of models appear in a catalogue kept up to date, which spells out how much platform markup each price carries. Once a run starts, its price is locked, so a change made afterwards never reaches back to it. Twigg publishes no subscription price.

Intended range

Twigg presents itself as suitable at every scale, whether a personal agent or an enterprise application. It describes its own offer as one API for LLM use with no provider lock-in.