Troubleshooting
Sign-in & Email
Verification or password-reset email never arrives
Rolv sends verification and password-reset mail from noreply@rolv.io (not Firebase’s default noreply@…firebaseapp.com).
- Check spam / junk, and wait a minute — rate limits apply
- On the verification screen, click Resend verification email, then I’ve verified after you open the link
- On sign-in, use Forgot password? again with the same address you registered
- Confirm the address is spelled correctly; the reset form always says “sent” so it does not reveal whether an account exists
Google sign-in does not need this flow.
Environment Setup
Mamba/Micromamba Not Found
If rolv can't create sandbox environments:
- Open Settings → Execution Tools
- Set the Mamba Executable Path to the full path (e.g.,
/opt/homebrew/bin/micromamba) - Save settings
If you don't have Mamba installed, install Micromamba:
"${SHELL}" <(curl -L micro.mamba.pm/install.sh)
Python/R Version Issues
If a specific Python or R version fails to create an environment:
- Ensure the version is available in your Mamba channels
- Try switching to a different version in Settings → Execution Tools
- After changing R version, start a new session so Mamba can build a new template (
_template_r461for 4.6.1, and so on) - Check that your internet connection is available (Mamba downloads packages on first use)
Environment Spec Errors (_template_py...)
If you see libmamba Could not solve for environment specs mentioning _template_:
- Open Terminal and delete cached template environments:
rm -rf ~/.rolv/envs/_template_* - Re-run your recipe. Rolv will automatically generate a fresh, clean environment.
Code Execution
Code Won't Run
- Chat only asks for a working directory when a disk tool needs one (not on every message). Use the folder icon in the header if none is set.
- Verify Auto-Run is enabled if you expect automatic execution. If it is off, the agent stops after suggesting code — run the block yourself or turn Auto-run back on.
- Look at the Logs tab in the node inspector for error details
- Use Stop in chat or Stop on the graph toolbar to cancel a stuck local script
Generate Data failed, or the inputs stayed empty
Open the Data Input node and expand Test data. The full steps are in Graph View — Test data.
- The editor is empty. Sign in, choose R, Python, or Bash, and click Starter. Starter needs a signed-in account so it can read the pipeline. If it cannot reach the model and the editor is still empty, Rolv inserts a short outline you can edit.
- Generate Data says generation failed. Read the message in the alert. Set the language picker to the language the script is written in, fix the script, and click Generate Data again. Generate Data installs libraries the script imports before it runs.
- Generate Data finishes and the input fields stay empty. The script has to write
test_input.jsonin the session folder at the end of that run. The names in that file are the names on the Data Input card. A file input’s value is the sample filename the script just created. A column name or a number is written there too. Atest_input.jsonleft from an earlier run is ignored. After a successful copy, the same values are saved asinput.json, which the pipeline reads. - Run with Test Data says there is no test script. Click Starter on the Data Input node first.
Recipe Data Input Became a Compute Node
Older Save as Recipe exports stored I/O as script nodes. Re-open the original graph and Save as Recipe again so Data Input/Output round-trip as I/O. Names, descriptions, and defaults are kept. Local file paths are cleared.
Use Recipe opened a different session
Clicking Use Recipe on a card (or Run Recipe in the library) always starts a new graph session. To load the recipe into the chat you are already in, type the recipe name or run it / yes instead. See Chat View — Use Recipe vs saying “run it”.
Sidebar opened Graph when I left Chat
Rolv now restores the last Chat or Graph view per session. If you still land in Graph, click ← Back to Chat once; the next sidebar click should return to Chat.
Missing Dependencies
When Python or R code fails with a missing package error:
- Rolv automatically normalizes package names (such as
cv2toopencv-pythonorsklearntoscikit-learn) and falls back topipif Conda lacks the package. - If a package still fails to import, the Node Inspector immediately flags the node and displays the installation logs.
- You can ask the agent in chat to install or adapt the package: "Install scikit-learn" or "Use another library".
- For community recipes, add the package name to the recipe's
env/files (python-requirements.txtorr-packages.txt).
SSH Execution Failures
If remote execution fails:
- Open Settings → Execution Tools → Remote Compute
- Confirm the target type is SSH (OpenStack and Slurm can be saved but do not run yet)
- Click Test on the connection
- Verify: host reachable, username correct, key/password valid
- Check that the remote Python command works (e.g.,
python3orconda run -n myenv python) - Ensure the remote work directory exists and is writable
If Add Target is disabled, you are at your remote-compute slot cap. SSH, OpenStack, and Slurm share that cap.
Search & Embeddings
Recipe Search Not Working
If semantic search returns poor results:
- Open Settings → API Hub → Embedding Model
- Click Test Connection to verify the model is reachable
- If you recently changed embedding models, click Re-index in the prompt that appears after saving
Re-indexing Takes Too Long
Re-indexing processes all local recipes sequentially. Progress is shown in the Settings panel. For large libraries, this may take a few minutes.
Cloud Sync
Merge Conflicts
When a Smart Merge produces conflicts:
- Open the recipe in the Designer
- Check the Changes tab for a diff view
- Manually resolve conflicting sections
- Save when satisfied
"Recipe No Longer on Cloud"
If a cloud recipe was removed by its maintainer, you'll see a "No Longer on Cloud" badge. Your local copy still works — it just won't receive future updates.
Token Limits
Quota Exceeded
When you hit your daily or monthly token limit:
- Wait for the daily reset (midnight UTC)
- Or configure a Custom LLM in Settings → API Hub (OpenAI / Groq / Ollama / OpenRouter presets). Localhost Ollama can leave the API key empty.
- Set the usage mode to "Always Use" or "Use as Fallback"
Credits Running Low
Monitor your credit balance in Settings → Token Usage. Heavy local simulations skip cloud credit costs. Credits purchase is coming soon.
Research nodes
The report stops in the middle of a table
The assistant often writes a long essay about the first gene or topic and runs out of room before the rest. Searches in the Logs tab can still show every gene. The markdown is still saved as {label}_report.md. For a table you will reuse, tick Structured output and list field names, or retrieve with a compute node first. Details: Graph View — Research nodes.
Structured output only searched one gene
By default, structured output gathers evidence in a single pass. If you want the node to iterate across every row, gene, or biomarker in your table, check Batch iteration across items (ForEach) in the node settings. Rolv will then run parallel queries across each item in your dataset (up to 100 items) and compile the results into both structured JSON and an aggregated report table.
Inspector asks “This will not search every item”
When Batch iteration across items (ForEach) is disabled and instructions ask to process a list of items, the inspector displays a reminder dialog. Enable Batch iteration across items (ForEach) to perform the full multi-item search, or click Run anyway for a single-pass summary.
Research failed with fetch failed
Unpackaged npm run dev talks to http://localhost:8080. Start the local gateway, or run ROLV_GATEWAY_URL=https://gateway.rolv.io npm run dev. Packaged builds use production.
I connected two nodes but the second cannot see the file
Arrows do not auto-pass spreadsheets. The first node must write a named file; the second must open that same name (in code, or in the research Instructions). See What the arrows between nodes mean.
App Updates
When a new version of rolv is available, it downloads in the background and a toast appears at the bottom of the screen:
macOS (unsigned builds):
- Click Show in Finder to open the downloaded update file, then install it manually (right-click → Open if Gatekeeper blocks it)
- Or click Download from Releases for the
.dmgon GitHub
Windows / Linux:
- Click Restart Now to install the downloaded update immediately
- Or click Download manually to get the installer from GitHub Releases
- Click ✕ to dismiss; a downloaded update still applies on the next restart
← Back to: Index