← All posts
Product7 min read

Thirteen connectors, signed webhooks, and a typed SDK

A big batch of platform work: nine new connectors, Stripe-style signed webhooks, the official TypeScript SDK, a durable run queue that survives deploys, 34 lending templates, and solution pages for six industries.

The summer went into making LoopLlama something you can build real systems on, not just run demos against. Here's what shipped.

Thirteen connectors#

Gmail, Salesforce, HubSpot, and MySQL were joined by Outlook, Slack, Twilio SMS, Plaid, DocuSign, Google Sheets, Google Calendar, PostgreSQL, and a generic authenticated HTTP API connector. Each connector exposes capabilities — gmail.read separately from gmail.send, salesforce.query separately from salesforce.update — that you enable per agent, and any send or write can require a human's approval before it happens. OAuth tokens and credentials are encrypted with a key that never lives in the database.

  • Plaid gives underwriting agents read-only access to bank transactions and balances, so the numbers come from the bank, not the model.
  • DocuSign and Twilio put e-signature and SMS in an agent's hands, behind approval gates.
  • The HTTP API connector wraps any REST service with a bearer token or API key, so internal tools work without a bespoke integration.

Signed webhooks#

Polling a run is fine for a script; it's the wrong shape for a production integration. Webhook endpoints now receive a signed JSON POST when a run starts, completes, fails, or pauses for a human. Signing follows the Stripe-style scheme — a LoopLlama-Signature header with a timestamp and an HMAC-SHA256 over the body — and the SDK verifies it for you. Delivery is retried with backoff, every attempt is logged in the dashboard, and an endpoint that keeps failing is auto-disabled so we never stall a run on a dead URL.

The TypeScript SDK#

@loopllama/sdk is the official typed client: workflows, runs, webhooks, a poll helper that waits for a terminal state, and a verifyWebhookSignature helper for your endpoint. The REST API is unchanged, so anything the SDK does you can still do with curl from any language.

A run queue that survives deploys#

Runs used to be fire-and-forget inside the web process, which meant a deploy mid-run could leave work stranded. Now every instance runs a worker that claims runs with a lease and renews it while the run executes. When a process dies — deploy, crash, out-of-memory — its lease expires and another worker resumes the run from the last completed step. Steps that could have performed side effects aren't replayed blindly: if the interrupted agent had access to tools that send or write, the run stops with a clear message instead of risking a duplicate email.

Templates and six industries#

The template gallery now holds 34 ready-made lending workflows — broker submission triage, stips collection, Plaid cash-flow underwriting, funding agreements through DocuSign, NSF follow-up, daily ops digests — each installable in one click and editable from there. Alongside lending, there are now solution pages for insurance agencies, property management, recruiting and staffing, accounting firms, and legal intake, each with the lifecycle, a worked example, and the crews teams start with.

Need something that isn't in the gallery? Describe the job in plain English and the AI workflow builder drafts the crew, grounded in your connected Salesforce schema. Or compose one from the Agent Hub's 30 reusable roles.

Written by The LoopLlama team.

Run your first agent crew in five minutes.

Get an API key and put these ideas to work. Pay only for the steps your agents run.