SkillZ ingests enterprise repositories, detects embedded AI workflows, extracts reusable execution Skills, and operationalizes AI execution with runtime observability.
Trusted by platform & AI teams at
The problem
Prompts live inside service code. Model calls are duplicated across teams. There is no consistent runtime, no central trace, no governance.
Critical prompts buried across hundreds of source files.
Each team rebuilds the same classification, extraction, routing logic.
Production AI calls are invisible to platform and ML teams.
Failures require deep log spelunking with no execution graph.
Retries, timeouts, fallbacks vary per service.
No registry, no versioning, no source of truth.
How SkillZ works
Connect GitHub, GitLab, Bitbucket.
Scan for prompts, model calls, tools.
Generate versioned Skill specs.
Execute through a standard runtime.
Trace every step, every retry.
Developer-first
Invoke any registered Skill through a stable, versioned API. Every call returns structured outputs, a full execution trace, and aggregated cost & latency metrics — no glue code.
import { SkillZ } from "@skillz/sdk";
const skillz = new SkillZ({ apiKey: process.env.SKILLZ_API_KEY });
const result = await skillz.skills.invoke("support.classify_ticket", {
input: {
subject: "Cannot reset my password",
body: "I've tried 3 times…",
},
trace: { tenant: "acme", session: "s-2c4a91" },
});
console.log(result.output.category); // "auth.password_reset"
console.log(result.trace.totalLatencyMs);Automatically detect AI workflows across your monorepo and services.
Identify embedded prompts, model calls, tools and execution logic.
Convert ad-hoc workflows into versioned, reusable Skill specs.
Execute every Skill through one structured runtime layer.
Trace every execution step. Inspect model calls, tools, retries.
Versioning, RBAC, audit logs, environment isolation.
Per-Skill budgets, model fallback chains, hard caps per tenant.
Encrypted secret storage with redaction policies on trace capture.
Promote, canary, and roll back Skill versions like deployments.
Use cases
A sample of Skills our customers run in production today.
Classify, route, and draft replies to tickets across Zendesk, Salesforce, and Intercom.
Extract structured data from invoices, contracts, KYC packets with provenance.
Score transactions and escalate edge-cases with explainable rationales.
Operationalize RAG over Notion, Confluence, Drive — with audit logs.
Automate reconciliations, exceptions, and approvals across ERPs.
Govern translation, tone, and compliance review at scale.
Integrations
SkillZ is gateway-agnostic and observability-native. Bring your own models, identity, and telemetry pipeline.
Operational trust
Inspect structured traces, runtime logs, and execution metrics for every Skill. Built on OpenTelemetry-compatible primitives.
FAQ
Static analysis identifies LLM/SDK call sites, prompt literals, and tool invocations. An LLM-assisted pass proposes a Skill spec, which an engineer reviews in the Studio before promotion.
No. Repository contents are scanned in ephemeral workers and discarded after Skill extraction. Only metadata, detected call-sites, and the resulting Skill specs are retained.
OpenAI, Anthropic, Azure OpenAI, Google Vertex, AWS Bedrock, Mistral, Cohere, plus self-hosted endpoints (vLLM, TGI, Ollama). Skills are model-agnostic; swap providers via config.
Yes. The Enterprise plan ships a Helm chart and Terraform module for AWS, GCP, and Azure. Runtime and control plane can be deployed in your VPC.
Yes — SOC 2 Type II. We also support PII redaction at trace capture, regional data residency, and BYOK encryption for secrets.
Per Skill execution, not per seat. Repository ingestion, observability storage, and SSO are included. Model-token costs pass through to your provider.
Get hands-on with the platform and see real workflows extracted from your repositories.