ReasFlow Paper Copilot

The ReasFlow Paper Copilot makes it extremely easy to orchestrate and execute a full scientific research and writing workflow in the ReasLab platform.

Accessing the Agent

Open the chat panel (make sure you have configured your LLM Provider in Settings first), and select ReasFlow Copilot from the dropdown to start your scientific writing journey.

Basic Features

The scientific writing assistant supports two interaction modes.

Mode 1 – Default (MetaAgent‑driven):
You directly converse with the main agent (MetaAgent). MetaAgent interprets your research intent, plans the workflow, and automatically calls the appropriate sub‑agents when needed. For example, if you say, “Please provide the convergence proof for algorithm A,” MetaAgent will first confirm the task with you and then invoke the ProverAgent to carry out the proof.

Mode 2 – Fine‑grained control (direct sub‑agent calls):
You can also talk directly to any dedicated sub‑agent using @ mentions. This gives you precise control over individual steps. The available sub‑agents are:

• SurveyAgent – via @survey / @survey-auto: conducts literature reviews and writes the "Related Work" sections.
• AlgorithmAgent – via @algorithm / @algorithm-auto: designs algorithms, provides pseudocode, and verifies the algorithm in Python.
• ProverAgent – via @prover / @prover-auto: establishes theoretical properties, provides assumptions, and conducts proofs.
• ExperimentAgent – via @experiment / @experiment-auto: designs and executes experiments, then reviews the results.
• IntroductionAgent – via @intro / @intro-auto: summarizes research progress and drafts the "Introduction" section.
• PaperAgent – via @paper: aggregates all research contents and drafts the final paper.

Note on -auto suffix:
By default, agents without the -auto suffix run in interactive confirmation mode (they ask for your confirmation before each step). Adding the -auto suffix enables fully automatic execution without interruption.

To return to the default MetaAgent‑driven mode at any time, simply type @meta.

For more detailed instructions, please refer to the detailed user manual.

API Configuration

ReasLab provides free built‑in API models (ReasPro, ReasProX) for all users. However, for better stability, we strongly recommend using your own API key. Currently, the ReasFlow system offers excellent support for the GPT series of models. To enhance citation quality, ReasLab also includes a free built‑in Semantic Scholar API. However, because the shared public API is subject to rate limits, we recommend that users apply for their own Semantic Scholar API key to avoid potential performance issues. You can request one here: Semantic Scholar API Key Application Form. Using a stable API (whether from your own key or a reliable provider) can resolve 80% of the issues you may encounter during use. This is why we recommend using your own API key when possible.

Example Workflow

To get the best results from the agent, the recommended workflow is:

1. Prompt Input: Provide an initial natural-language prompt describing the research or paper-writing task.
2. Plan Generation: The Copilot first returns a preliminary research plan.
3. User Feedback: You can request detailed adjustments, or type Continue to let the agent proceed.
4. Iterative Execution: The Copilot carries out the subtasks according to the plan, and after each stage it provides the generated content together with a summary of the next step.
5. Manual Correction: If part of the generated pipeline or subsystem is not meeting expectations, you can explicitly invoke a specific @ command such as @experiment to focus on and improve that area.
6. Final Output: The Copilot gathers all content and generates the final PDF version of the paper.

Sample Project

You can refer to the following example project to see the expected deliverable: Scientific Paper Writing

Example Video

Watch the full scenario demo: ReasFlow Paper Copilot.

Trouble Shooting

1. API errors – First, try continuing the conversation with ReasFlow to trigger an automatic retry.
2. Persistent errors – If the error continues, start a new conversation session and retry.
3. Sub-agent internal errors – If MetaAgent reports that a sub-agent keeps returning an internal error, instruct MetaAgent to retry using a new session for that sub-agent.
4. New session doesn't resolve the error – Try copying the project and continue your task in the new project.
5. If all above steps fail – Feel free to join our WeChat or Feishu group for support (scan the QR codes below), or email us at [email protected]. Sharing your project's conversation history will help us diagnose the issue more effectively.

WeChat Group

Feishu Group

FAQ

No FAQs yet. This section will be updated regularly. Please stay tuned.