Skip to content
Rivul AI
Explore
FeaturesRivul MCPToolsGuides
PricingDocsBlogAbout
Sign in
FeaturesPricingRivul MCPDocsBlogToolsGuidesAbout
Rivul AI

A free multimodal AI research workspace for papers, documents and recordings.

Explore

FeaturesRivul MCP

Resources

DocumentationResearch guidesFree citation toolsBlogAll resources
About Rivul AIContact and support
© 2026 Rivul AI
PrivacyTerms

Overview
  • Get started with Rivul AI
  • Research workflows
  • Rivul AI on a phone
  • Search papers
  • Library and the Add menu
  • Document OCR
  • Transcribe
  • Listen
  • Connectors
  • Ask Rivul
  • Research
  • Agents and Runs
  • Skills
  • What is MCP for AI research?
  • Connect Claude Code to Rivul MCP
  • AI research with OpenAI Codex and Rivul MCP
  • Connect Cursor to Rivul MCP
  • Connect Google Antigravity to Rivul MCP
  • Connect Meta Muse Code to Rivul MCP
  • Rivul MCP tool reference
  • Rivul MCP usage and fair-use limits
  • Citation tools
  • Citations
  • Writing
  • Analyze
  • Claim checking
  • Review
  • Export
  • Settings, sign-in and your AI key
  • Privacy and AI use
  1. Documentation
  2. Agents
  3. Agents and Runs

Agents and Runs

Research agents search papers and your library, read the most relevant ones, and write a cited report with your AI provider. Runs keeps every run.

On this page
  • Before you start
  • Create an agent
  • Run an agent
  • Runs
  • Limits
Rivul AI Research agents page for a new account, with Run history, Create agent, and a note that agents need your own AI provider connected in Settings.
Agents in a new account. Create agent stays unavailable until an AI provider is connected, because every agent step runs on your own key.

Before you start

Agents reason with your own AI provider. Connect one in Settings → AI provider; until then the Agents page says so and Create agent is unavailable. Your provider bills you for each model step.

Create an agent

1

Step 1: Open Agents in the sidebar, under Agents, and click Create agent.

2

Step 2: Give it a Name and a Goal: what the agent looks for, and for whom. Instructions are optional, for method or style.

3

Step 3: Pick its Sources: Rivul Index, My library, and Web where your AI provider offers web search (Google Gemini, OpenAI, Anthropic, and xAI).

4

Step 4: Pick a Depth: Quick runs take up to 5 model steps and 3 web searches, and Thorough runs up to 12 steps and 6 web searches.

5

Step 5: Pick an Output: Report, Evidence table, Annotated bibliography, or Related work section.

6

Step 6: Attach up to 4 Skills to guide its method and writing, then click Create.

Run an agent

Type a Research question and click Run. The run shows its steps as they happen: each search, each paper read, and the report being written. You can leave the page; the run keeps going and appears in Runs.

The report cites each claim with a source key that maps to a record the run actually read. A skill changes how the agent works and writes; it never adds a tool or a source.

Runs

Agent runs (rivul.ai/app/runs) lists every run with its status. Open one to see its Steps, the sources it read, and the Report. Copy the report, or click Download .md for Markdown with a source list built from the run's own records.

Check each source before you cite it. A run that stopped or failed keeps the steps and sources it reached.

Limits

Each account keeps up to 20 agents and can start up to 30 runs a day. A run also has a token budget on your key, so a run whose searches return long passages stops searching and writes its report sooner.

The Web source uses your AI provider's own web search, which receives only your question and the titles of Rivul Index papers, never your library text.

  • Skills
  • Privacy and AI use
PreviousResearchNextSkills

Something here wrong or missing? Use the contact form.

On this page

  • Before you start
  • Create an agent
  • Run an agent
  • Runs
  • Limits
Rivul AI
Explore
FeaturesRivul MCPToolsGuides
PricingDocsBlogAbout
Sign in
FeaturesPricingRivul MCPDocsBlogToolsGuidesAbout