Hookdeck

📹 Webhooks by Example Episode 2: GitLab webhooks as a virtual pet for repo health. We sat down with Colleen Lake, Developer Advocate at GitLab, to walk through GitLab's webhook platform using her Tanukigotchi demo: a dashboard pet whose mood tracks pushes, merge requests, pipelines, and work items. The walkthrough covers: → Event triggers, branch filters, custom templates and headers → Signing tokens and why verification is not optional → Emoji events: a webhook fired by a reaction, and what you can gate with one → GitLab Duo building the emoji handler live: draft MR, pipeline, human review, merge, deploy, tested on camera Episode 1 covered consuming webhooks in production. Episode 2 is the producer side, from the team that builds the platform. Read the write-up: https://lnkd.in/eCnP8YhM Watch Episode 2: https://lnkd.in/ew_i6Vry

Hookdeck CLI v3 is out: AI agents can now retry failed webhooks and update connections in Hookdeck's Event Gateway. The Hookdeck CLI forwards webhooks to your local server, manages your Event Gateway and Outpost resources, and runs MCP servers so agents can work with them. In v3: → Agents can create, update and delete connections, sources and destinations, retry or cancel events, and dismiss issues. Write tools only exist when you start the MCP server with --allow-write. → Reads and writes are separate tools. MCP clients grant permission per tool name, so you can let an agent investigate freely and approve each change. → New hookdeck outpost commands and an Outpost MCP server. Outpost sends webhooks and events to your users' destinations. Already using the MCP server? Tool names have changed, so update any tools you pre-approved in your client. Read: https://hkdk.link/cli-v3

One tenant's traffic burst becomes everyone's backlog. Until now... Today we're launching Delivery Groups. It's the only managed solution for fair multi-tenant webhook delivery that lives in the delivery path. No consumer rewrite needed. Set a key from your webhook payload: shop_id, account_id, whatever maps to "tenant" in the body, headers, query, or path. Each unique value dynamically gets its own delivery group with its own rate. While you get observability to know what's going on. Use it to stop noisy neighbor issues, or stay within downstream API rate limits. Alexandre Bouchard shows it in action:

Verify, queue, deduplicate, retry, and rate-limit webhooks that trigger your n8n workflows. The Hookdeck Event Gateway node for n8n is here, bolstering reliability for your webhook-triggered workflows. Two nodes: - Hookdeck Event Gateway Trigger replaces the plain Webhook trigger. When you activate your workflow, the node provisions a Hookdeck connection pointing at that workflow and hands you a public source URL to give your provider. Every event now flows through Event Gateway first: verified, queued durably, deduplicated, rate-limited, and retried on failure. - Hookdeck Event Gateway is the action node. It works with events, delivery attempts, requests, issues, sources, destinations, and connections, so you can retry a failed event or pull up the original request a provider sent. Combine these nodes with n8n's own AI agent to create smart, efficient workflows. Get started: Install on n8n Cloud or self-hosted. Search 'Hookdeck' from the nodes panel, or install the package: @hookdeck/n8n-nodes-hookdeck

🆕 Extended destination timeouts up to 15 minutes are now available in Event Gateway. Ideal for AI tasks and heavy backend workloads, extended timeouts let your destination server process the entire event lifecycle synchronously. This eliminates the need for polling or internal queueing, while giving your handlers the buffer to work around cold starts. You can change this under 'Advanced configuration' in Settings for each destination, with a default of 1 minute.

We built Hookdeck Evals to hunt out the friction between your agent and Hookdeck. We want every Hookdeck product to work for an AI agent as well as it works for you. Hand Claude Code or Codex a Hookdeck project and ask it to receive Stripe webhooks, and it should get there without you stepping in. Mostly it does. Sometimes it doesn't. The problem, though, is how we find out. When a person hits a docs page that is wrong, some of them tell us. Agents don't. An agent that can't find what it needs guesses, works around the gap, and will probably still report success. Nobody files a bug because nobody knows there was one. So we built Hookdeck Evals to go looking for that friction. 19 scenarios, multiple models, run in live Hookdeck projects and scored on what the agent built rather than on what it said it built. Every finding becomes a GitHub issue, whether the fix belongs in the docs, the API, the CLI, our skills, or the product itself. If an agent has gone wrong on Hookdeck, we want to hear about it. The gaps we haven't found are the ones we need most. Check out https://hookdeck.com/evals to learn more.

Grok bot routines can now be woken by a webhook. Create a routine, add a trigger, pick Webhook, and you get a URL and a key. Stripe, Shopify, Notion, a sensor, a GitHub Action: anything that can POST can start a run. It's fun to play with. It's even better with Hookdeck's Event Gateway in front, because three things become obvious the moment you connect a real provider: - The body reaches the routine verbatim. A raw Shopify order is several hundred lines of JSON for the Bot to work through. Shape it with a transformation first and send the six fields the routine needs. - Every request is a run. Providers don't know you only care about `payment_intent.payment_failed`. Filter on the connection so only the right payloads wake the Bot. - Providers retry. A slow response turns one event into three runs and three Slack messages. De-dupe so one event is one run. We tested the triggers end to end. Check out our guide on how to set the whole thing up: https://lnkd.in/eKDKiX6J

We've released hermes-hookdeck, a plugin that puts Hookdeck's Event Gateway in front of Nous Research Hermes for webhook reliability. - Traffic bursts get queued - Events are deduplicated and survive restarts - Failed runs get redelivered - In-flight events are recovered at boot or crash - Signature verification for 140+ provider schemes And all with no public port/URL required. The agent can also triage its own failures, with tools to list failed events and bulk retry them, alongside status, pause, resume, retry and doctor commands for whoever is running it. Works with a free Hookdeck account. Install via Homebrew or npm and run `hermes hookdeck setup`.

New in Event Gateway: Request replay Replay re-ingests the original request through the full pipeline as if it had just arrived — creating a new request and a new event for every target connection. By default it fans out to every active connection on the request's source, including ones you've added since. You can also narrow the replay to specific connections, or point it at a different source entirely. What that gives you: • Backfill a new connection with historical events • Reprocess requests after changing a transformation or filter • Reroute traffic that landed in the wrong place • Test against real production data