Pipeline failed

What this means

The pipeline reached the Failed state. The agents stopped trying and didn't produce a merge request, or the merge request was opened but the run halted before it could merge.

A failed pipeline does not retry on its own. Bilbis records what happened and waits for you to look at it.

When to use it

The state badge on a pipeline is red and reads Failed.
The pipelines list shows a row in the Failed filter.
The dashboard's recent pipelines list flags one as Failed.

Before you start

Open the pipeline detail page. Useful tabs:

Diagnostics - Bilbis's internal view. Almost always names the reason.
Timeline - what happened, in order. The last few rows tell the story.
CI runs - when CI was involved.
LLM calls - when cost is the suspect.

Steps

1. Read Diagnostics first

Click the Diagnostics tab on the pipeline detail page. The reason is usually one of these:

Reason on Diagnostics	What it means	Where to go next
Budget cap reached	The pipeline halted because cost exceeded the cap.	Raise the cap on a fresh run, or pick a cheaper engine. See Budgets, dry runs, and priority.
LLM provider rejected the request	Your LLM credential is invalid, expired, or the provider returned an error.	Check Settings → Integrations.
LLM provider over quota	Your LLM key hit a per-account or per-key rate limit.	Check the provider's dashboard. Wait or top up.
CI never converged	The agent tried multiple CI fixes and the failure stayed.	See CI still failing.
No config found for repo	A known backend gap. The repo is registered but the dispatcher can't read its config.	See No config found for repo.
Push rejected by provider	Your Git credential lacks permission, or the branch is protected.	Check the credential scopes and branch protection rules.
Merge rejected by provider	The MR's target branch is protected and required approvals are missing.	Approve through the right channel and re-dispatch.
Internal error	Something on the Bilbis side went wrong.	Note the pipeline id and report it.

If Diagnostics is empty or unclear, scroll the Timeline tab. The last few rows usually have an escalation row with a "Reason" callout.

2. Decide whether to retry

Bilbis does not retry automatically. To run the same task again:

Open Pipelines → New.
Type or paste the same task description, or pick a saved template.
Address whatever caused the failure (raise the budget, fix the credential, clarify the task).
Dispatch.

There's no "retry" button by design - most failures need a tweak before another run is worth doing.

3. Decide what to do with the MR, if any

If the failure happened after the MR was opened, the MR may still be on your provider with the agent's branch pushed. Three options:

Continue manually. Pull the branch and finish the work yourself.
Close the MR and re-dispatch. Useful if the agent's diff was off-track.
Approve manually on the provider. If the failure was on Bilbis's side but the diff is good, you can merge from GitLab or GitHub directly. The pipeline stays Failed in Bilbis, but the work lands.

4. Check the budget

A run that failed early still spent money. Open the LLM calls tab and the header to see the cost. Adjust the budget cap on the next run if needed.

Common causes by tab

What you see	Likely cause
Cost in header is at or near the budget cap	Budget halt.
CI runs tab shows red checks	CI failure.
Clarifications tab is empty but state is Failed	Agent didn't have a question - the failure was elsewhere.
LLM calls tab shows a 401 or 429	LLM credential issue.
Timeline has an escalation row with a long reason string	Agent gave up and surfaced a structured reason.

What does not help

Refreshing the page. The pipeline is terminal. State doesn't change after Failed.
Cancelling. Cancel only stops further state transitions; a Failed pipeline is already past that point.
Editing the pipeline. Pipelines aren't editable. Dispatch a new one.

Permissions

Action	Who can do it
Read the failure reason	Any role.
Re-dispatch a fresh pipeline	Any role with permission to create pipelines.
Approve or merge directly on the provider	Subject to your Git provider's permissions.

Pipeline troubleshooting (overview) - symptom-based jump table.
Pipeline states - every state and badge color.
Review pipeline results - read the detail page tab by tab.
No config found for repo - common specific cause.
CI still failing - common specific cause.
Awaiting clarification - different shape of pause that can convert into a Failed run if not answered.

On this page