Research, on your terms

Your question.
Your model. Your workflow.

Try a setup here. Take the whole session into your work.

Open session

Checking the local research connection…

Run this setup locally

Export your session, then use your local Grainulator installation. Your provider credentials stay on your machine.

First time here? Set up the local build. Use the GitHub release; this version is not published to npm.

grainulator preview

Open the local playground and choose Open session, or continue in your terminal:

grainulator research --session session.json --dir ./research-session

From a source checkout, use node bin/grainulator.js in place of grainulator. Configure OPENAI_API_KEY or OPENROUTER_API_KEY locally, or enter your key in the local playground.

01 / Set up your research Research question

What should Grainulator add?

Tune the work around your model. Open ⓘ to see how each feature works.

Model connection

Provider

Model ID

Astra is the starting model. You can choose another model ID supported by your API account.

API key · optional if configured locally

Kept in this tab. The key field is excluded from exports.

Uses OPENAI_API_KEY from your local server when this field is empty.

Search the web · OpenAI only

Search is available to the model. Citations returned by the provider appear with each pass.

Context & instructions

What should the model know?

Include source text when possible. URLs alone are not fetched when web search is off.

What should it prioritize?

Set priorities, comparison criteria, or what needs independent verification.

Limits & reasoning

Reasoning effort

Output tokens / pass

Timeout / pass (seconds)

Retries / pass

Provider charges apply, including web search. These limits bound each pass; they are not a dollar budget. A higher output limit leaves room for reasoning.

Your question, context, and prior passes are sent to the provider you select. How data is handled

02 / Follow the work

From question to decision.

Ready

Research
Challenge
Blind spots
Claims
Synthesize

A workflow you can inspect.

Watch each pass arrive. Read the findings, inspect sources, and see how the recommendation develops.

Your session keeps the model settings, context, and completed passes together. You can stop here and continue locally.

The example shows a blocked evidence gate. It makes no API calls.

03 / Take it with you

Keep the work. Keep going.

Export a session to reopen here or resume from the CLI. Export a brief to carry the context into your agent.

node bin/grainulator.js research \
  --session session.json --dir ./research-session

Uses your local provider key. Completed passes are reused; editing the question or settings starts a new run.

For an authenticated compatible endpoint, set a local environment variable such as MODEL_API_KEY and add --api-key-env MODEL_API_KEY to the CLI command. Use the variable name, never the key itself. Exported sessions do not carry credentials.

Your question.Your model. Your workflow.

What should Grainulator add?

From question to decision.

A workflow you can inspect.

Keep the work. Keep going.

Your question.
Your model. Your workflow.