Documentation · v2026.05

Build with Quellra.

Welcome. This is the developer documentation for Quellra — the AI customer support platform that resolves issues end-to-end. Start with the quickstart below, then explore concepts, agents, and the workflow DSL.

Quickstart

Install the SDK and run your first workflow against a sandbox tenant.

terminalbash

# install
npm install @quellra/sdk

# or
pip install quellra

# authenticate
export QUELLRA_API_KEY=rsv_live_********

Run a sample workflow:

recover-failed-payments.tsTypeScript

import { Quellra } from "@quellra/sdk";

const rsv = new Quellra();

await rsv.workflows.run("recover_failed_payments", {
  customer_id: "cus_3PqA…",
  context:     { plan: "growth" }
});

Core concepts

Quellra organizes work around five primitives:

Workflow — a named, versioned DAG of steps with a trigger and an outcome.
Agent — a specialized role (planner, executor, verifier, writer) with a sandboxed toolset.
Tool — a typed callable (REST, MCP, custom function) with per-action authz.
Trace — the immutable record of one workflow run: every step, tool call, model call, decision.
Eval — a golden case + grader that runs on every PR and in production shadow mode.

Agent architecture

Each workflow run instantiates one or more agents. The default loop is planner → retriever → executor → verifier → writer. Custom topologies are supported and can be visualised in the console.

The workflow DSL

Quellra workflows can be written in TypeScript, Python, or our declarative .rsv DSL. The DSL compiles down to the same DAG.

Triggers

Workflows can be invoked by events (telemetry, webhooks), schedules (cron), conversations (Slack, email, in-app), or anomaly thresholds. See Features → Triggers.

Tool registry

Quellra ships 200+ first-party tools. Register custom tools via the SDK or expose them over MCP. Each tool declares an input schema, output schema, side-effect class, and authorization scope.

Evals & CI

Every workflow has a companion eval set. PRs that drop accuracy below threshold fail CI. Production shadow mode replays a sample of real traces nightly to catch regressions before customers do.

Deployment

Quellra is delivered as managed multi-tenant SaaS, single-tenant VPC, on-prem, and BYOC. All deployments use the same control plane and APIs.

RBAC & SSO

Map your IdP groups to Quellra roles. Define per-action permissions. Required for Growth+ tiers.

Audit log

Every consequential action is appended to a per-tenant, immutable, hash-chained audit log. Export to SIEM or query via the console.

SIEM export

Streaming export to Splunk, Datadog, Elastic, or any S3-compatible store. Schema is documented and stable across minor versions.