Your best way ofworking with AI.

Turn your team's expertise into reusable workflows for AI agents. Assign tasks, guide the work, and review results in one place.

View on GitHub

  1. Build your process

    Turn your team's expertise into reusable workflows.

  2. Share your workflows

    Let teammates reuse the workflows you've saved.

  3. Make it better

    Bring what you learn into the next run.

NS-142

Add retry handling for failed deliveries

Mark as done
Add retries for failed webhook deliveries. Use exponential backoff and keep the delivery history on the task.

I'll check how deliveries are queued, then add retry handling and tests for failed requests.

3 tool calls
Read delivery workersrc/workers/deliver-webhook.ts
Read queue configurationsrc/queues/webhooks.ts
Update retry policysrc/workers/deliver-webhook.ts +38 −6

Failed deliveries now retry up to five times with exponential backoff. Each attempt records its response and timestamp in the delivery history.

  • Retries stop after a successful delivery
  • Failed attempts remain in delivery history
  • All 18 delivery tests pass
Work locallymain
$4.71

Your team's software factory

Turn your team's expertise into a shared process for building software with AI. Bring agents, skills, and workflows together so teammates can put that process to work and improve it across projects.

Turn a proven process into a repeatable workflow

Define how work moves from planning to review, with agent instructions, automated checks, and human approvals along the way. Save the workflow so your team can use it on the next task.

Stage
EngineeringGPT-6 Astra
ReviewParallel
Implement

Add exponential backoff for failed webhook deliveries. Keep the delivery history on the task.

Act
Review before continuing
Verify
pnpm test:once
Timeout: 10mStop on failure

Review retry limits, idempotency and delivery history. Leave findings on the task.

GPT-6 AstraPlan

Check webhook signatures and error handling for security issues.

Claude CodePlan

Stop making every team start from scratch

Turn what works for your most successful teams into a way of working others can adopt. Share agents, skills, and workflows that cover how you plan, build, test, and review software with AI. Give teams a proven starting point they can adapt and improve together.

/review-webhooks

Engineering

Review webhook changes for safe retries and verified deliveries.

Before approving a webhook change:

  1. Verify signatures before processing a delivery.
  2. Check retry limits and exponential backoff.
  3. Confirm duplicate deliveries are handled safely.

Record concrete findings on the task, with file references.

Files

references/review-checklist.md
Content

Review changes against the task’s acceptance criteria.

Report concrete findings with file references. Explain what needs to change and why.

Use the review-webhooks Skill when reviewing webhook changes.

See which agents need your attention.

Watch Tower brings your runs across an organization's projects into one place. Find unanswered questions, failures and work ready for review. Switch between List and Board views and triage runs in batches.

Watch Tower

Layout
Sections

Waiting for your input

Agents asked a question and are blocked until you answer.
ThreadStatusChangesUpdated
Review the webhook delivery planNorthstar·plan/webhook-delivery·NS-139Waiting for input+24−318m
Document webhook authenticationDocumentation·docs/webhook-auth·DOC-28Waiting for input+48−24m

Failed

Runs that stopped with an error and can be retried.
ThreadStatusChangesUpdated
Fix the signup form validationWebsite·fix/signup-validation·WEB-41Failed+18−68m

Needs your review

Finished threads you haven't marked as done yet.
ThreadStatusChangesUpdated
Add retry handling for failed deliveriesNorthstar·fix/delivery-retries·NS-142Needs review+86−122m
Keep delivery filters in the URLNorthstar·feat/delivery-filters·NS-136Needs review+42−81h

Review the work on the task

An agent writes a plan. You comment on a passage. The agent reads your feedback and revises it. Keep that review beside the task, along with editable Markdown documents, HTML reports and interactive outputs.

Implementation plan

Webhook delivery retries

Keep deliveries reliable when an endpoint is temporarily unavailable.

Retry policy

Retry failed deliveries up to three times with exponential backoff. Keep the delivery ID unchanged across attempts.

  • Retry timeouts and 5xx responses.
  • Record each attempt with its response status.
  • Stop after a successful delivery.
Verification

Test a failed attempt followed by a successful retry. Confirm the receiver processes the event once.

Keep related conversations together

Group related agent threads under one item and move between them in tabs. Share a thread or group when you want project teammates with thread access to follow along. Shared conversations are read-only for teammates.

Add retries for webhook deliveries. Keep the delivery ID stable across attempts.

The retry path now preserves the delivery ID and records each attempt. Timeouts and 5xx responses use exponential backoff.

The retry tests pass, including a failed attempt followed by a successful delivery.

Work locallymain
$4.71

Let agents work together

An agent can start threads for other agents and send them messages. One drafts the plan, another reviews the change, and their replies come back to the thread that asked. Each message links to the thread that sent it.

  • PlannerPlan retry handlingSent the plan
  • Docs WriterDocument retry behaviorWriting
NS-142

Add retry handling for failed deliveries

Add retries for failed webhook deliveries. Get the plan and the change reviewed.
Started threadPlan retry handling
Started threadReview retry handling
Plan: retry timeouts and 5xx responses with exponential backoff, keep the delivery ID stable and honor Retry-After on 429.
6m

Implemented the plan and sent the diff for review.

Two issues: the backoff has no jitter, and the attempt count resets when the worker restarts.
2m

Fixed both and asked the reviewer to check again.

  • ReviewerReview retry handlingSent 2 issues
  • TesterTest retry limitsTests pass

Check on your coding agent from your phone.

Your agent works on your code on a computer or server you connect to Convoy. Open Convoy in your phone's browser to see what the agent has done, answer its questions or ask for changes.

NS-142

Add webhook retries

Add retries for failed webhook deliveries.

I added exponential backoff for failed deliveries. Each retry keeps the same delivery ID so receivers can ignore duplicates.

Cap retries at three attempts.

I’ll cap each delivery at three attempts and test the limit.

1 tool call
Run command
pnpm test webhooks

✓ retries failed deliveries
✓ stops after three attempts
✓ preserves the delivery ID

3 tests passed

Done. Deliveries stop after the third attempt. All three retry tests pass.

Work locallymain
$4.71

Your process.
Your coding agents.

Use Claude Code, Codex, OpenCode, or Cursor with your team's workflows. Follow their work and review results in Convoy while agents run on your connected machines.

Claude Code
Codex
opencode
Cursor

Tasks and agents in one place.

Organize the work, assign an agent, and keep its progress and results connected to the task.

Assign a task to an agent

Choose an agent with the instructions and preferences for the job. Start a thread from the task, send follow-up instructions, and follow the run from your board.

NS-145

Add webhook retries

ListBacklog
TypeFeature
StatusTo Do
PriorityHigh
Assignee
AI Agents
People
Threads
  • Implement retriesnow
  • Review retry limitsnow
Due dateOct 16, 2026
Tagswebhooks
Createdby Alex Morgan 2 days ago

Description

Retry failed webhook deliveries with exponential backoff. Keep the delivery ID stable and stop after three attempts.

Keep tasks, files and pull requests together.

Plan work in Lists, break it into subtasks, and switch between Board and Table views. Keep priorities, comments, attachments and related pull requests on the task so the next step has its context.

Northstar

AllA

To Do

3

In Progress

2

In Review

2

Done

2

Keep plans and results with the task

Save plans, reviews and handoff notes as task artifacts. Teammates can review the documents, leave feedback and return to the decisions as the work continues.

NS-142

Retry failed webhook deliveries

Resources

Implementation planSaved

Webhook delivery retries

Keep deliveries reliable when an endpoint is temporarily unavailable.

Retry policy

Retry failed deliveries up to three times with exponential backoff. Keep the delivery ID unchanged across attempts.

  • Retry timeouts and 5xx responses.
  • Respect Retry-After on 429 responses.
  • Stop after a successful delivery.

Manage where your agents run.

Use convoy tui to manage local and Docker environments from your terminal. See connected projects, check service status, and add environments with keyboard shortcuts.

convoy tui
convoy   sandbox / docker2 registered projects
docker exec convoy-sandbox /home/node/.convoy/service/control.sock
ENVIRONMENTS
+ Add environment
▸ Local ● unix
▾ sandbox ● docker
All projects
+ Add project
Web app (web)
API (api)
[d] Details[a] Actions[c] Credentials
[s] Inspect status  [f] Retry connection
Connection:
connected
Service:
running   PID: 10444
Configuration:
valid
Service Node:
v22.3.0

Project configuration was captured when attached.

web: online, active 0
Project: Web app
Concurrency: default   Isolation: default
api: online, active 0
Project: API
Concurrency: default   Isolation: default
[←/→] panes  [↑/↓] select  [Enter] activate
● connectedsandbox · docker0 observed active

Run Convoy yourself.

Host Convoy's web app and backend on your own infrastructure. Connect the machines where your agents work and manage them from your own Convoy instance.

packtrain/convoy
Your teamPlan tasks, follow threads and answer agents in the browser
Live updates

Your infrastructure

Web appConvoy's Next.js app
Convex backendTasks, threads, artifacts and MCP
convoy connectoutbound connection
  • Your laptopmacOS
  • Build serverLinux
  • Cloud VMLinux
model API calls

Feature index

Agent execution

Agent threads

Follow coding agents on your connected machines. Read their responses, inspect tool calls and diffs, and send follow-up instructions in a thread.

Supported runtimes

Run Claude Code, Codex, OpenCode or Cursor, and pick the runtime and model for each thread. Steering and approvals depend on what each runtime supports.

Watch Tower

See agent runs from every project in your organization. Find questions, failures and work ready for review in List or Board view, with batch actions for triage.

Agent dispatch

Assign a task to an agent with its own instructions and preferences. Start a thread from the task and follow its progress from the board.

Steering & approvals

Review diffs and send feedback to the agent. Supported runtimes also let you steer a running agent and approve or deny tool use.

Branching

Branch a thread from any earlier run to try a different approach, or retry a run that failed.

Worktree isolation

Start a thread in a separate Git worktree to keep its changes apart from your main checkout.

Sandboxed agents

Run Claude Code or Codex in an OS sandbox (Seatbelt on macOS, bubblewrap on Linux), or inside the devcontainer the CLI generates.

Workflows

Build reusable stages with agent prompts, Bash checks and human approvals. Run lanes in parallel, follow each step, and retry or skip a failed step.

Prompt catalogs

Save reusable messages at organization or project level and insert them into a thread with a slash command.

Skills

Save procedures as reusable instruction bundles. Install organization templates into projects, adapt the copies, and call on them from a thread.

Reusable agents

Give an agent standing instructions for a role such as reviewer or implementer. Install organization templates or copy an agent from another project.

Thread groups

Keep related agent conversations together in one group, with a tab for each thread.

Thread sharing

Share a thread or thread group with project teammates who have thread access. They can follow the live conversation in read-only mode.

Scheduled runs/soon

Run agents on a schedule, such as nightly dependency bumps or weekly triage. Each run shows up as a task on the board with its full history.

The agent workforce

Usage & costs

See daily spend, tokens by model, and spend per member and per CLI for each project, plus the cost of each task and thread.

Fleet analytics/soon

See success rate and time to merge for each agent.

Agent pools & cloud execution/soon

Register your own machines or managed sandboxes as executors. Convoy spreads dispatched tasks across them so the queue keeps moving.

Desktop app/soon

A menu bar app that keeps your agents connected, so you don't need a terminal open.

Documents & review

Task artifacts

Keep plans, reviews and handoff notes on the task. Edit Markdown documents or open HTML reports and interactive outputs.

Artifact comments

Comment on a passage in a Markdown plan or review. Ask an agent to address the feedback, revise the document and resolve the comments.

Project memory/soon

Each new agent run starts with the decisions, specs, review comments and failed attempts from earlier work.

Project management

Real-time board

Organize work in Lists with Board and Table views. Set priorities, types and labels, choose which built-in statuses to use, and find tasks by refs such as ABC-123.

Rich tasks

Break work into subtasks and keep comments, attachments and artifacts with the task. Commits and pull requests linked by agents or the CLI show up on the task.

Keyboard-first tasks

Create, indent, reorder and update tasks from the keyboard, and open any task by its ref. Bulk edits, archive and delete come with undo.

Teams & permissions

Invite teammates to organizations and projects, and give each role the access it needs.

Git-forge integration/soon

Review comments go back to the agent, and merging closes the task.

Open & extensible

Code on GitHub

Read exactly what runs between you and your agents. Licensed under the Elastic License 2.0.

Self-hostable

Run Convoy's web app and backend on your own infrastructure.

MCP server

AI tools can read and update tasks, artifacts, comments, skills and prompts over MCP, and start agent threads. Claude Code connects with a browser sign-in, no API key needed.

Environment management

Manage local and Docker environments with convoy tui. Attach projects, inspect service status, and retry connections from a keyboard-driven terminal interface.

CLI

convoy connect links your agents to threads, and convoy tasks lets you script task work.

Enterprise & on-prem/soon

Air-gapped deployments, SSO with SAML, and audit logs that record every agent action and approval.

Build your team's way of working with AI.

Start with one workflow. Give your team a process they can reuse and improve.

View on GitHub