Teach the work you repeat.

You use Toolving to reach Conceiver. Models like Jev provide a fast System One response. Conceiver works alongside your LLM as System Two: it recalls and weighs evidence, learns over time, and shows where its decisions came from.

Conceiver starts with a Unicode lexicon and works with natural language. You teach it with input and expected-output examples, ask it questions, record what happened, and save its training sequence for reuse.

Start with the quickstart · Get an API key · SDK releases

How you work with it

Step	What you do
Create	Open an instance for your account.
Teach	Provide an input and the output Conceiver should learn.
Query	Receive an answer or an unresolved result with relevant evidence.
Resolve	Record what actually happened for a prediction.
Save	Keep an ordered training sequence of what your account learned.
Restore	Teach that sequence to a fresh Conceiver and recover its capabilities.

Instances in one account share that account's learning. Other accounts cannot access it. With storage, memories and learned capabilities are saved as an artifact that a fresh session can load. Without storage, an instance and its episodic memories are ephemeral and disappear when the instance retires.

Where to go next

Concepts: how instances, sequences, answers, evidence and outcomes fit together.
Python SDK: how you create, teach, ask, save and restore from Python.
HTTP API: routes, request bodies and interrupted-request recovery.
Examples: a complete teaching and reuse example.
Keys, credits and plans: keys, usage prices and storage.
Demo studio: interactive learning and evidence examples.
Chess research model: optional starter artifact with source and supplied-rule disclosure.

What you can use

Conceiver is available through the Toolving API and SDK at the URL shown in your account. Its sparse starting vocabulary leaves room for the language and transformations your workflows need.

Memory-enabled queries add contextual evidence and episodic recall. Support for raw JSON syntax is in active development. A JSON-encoded answer can still be carried as text and decoded by the SDK.