initlabs Atlas

Serval (serval.com)

9 min read

Serval

US-based, AI-native ITSM startup. Highest-funded direct competitor to initlabs in the same wedge. Frequently positioned by investors as the candidate “next ServiceNow.”

Snapshot

  • Category: AI-Native ITSM / “AI-native System of Record” / Enterprise Service Management
  • Tagline: “AI to give IT superpowers” — and (Series B framing) the next-era system of record
  • Front door: Slack, Microsoft Teams, email (with DKIM trust + guest handling), web portal (app.serval.com), phone (Twilio), and MCP (Claude.ai / Claude Desktop / any MCP client)
  • Compliance: SOC 2 Type II certified (confirmed in docs); TLS 1.3 in transit, AES-256 at rest; SAML SSO via WorkOS; SCIM; domain allowlist + enforcement; time-bound support tokens. Trust Center at trust.serval.com
  • Deployment: Cloud, Hybrid Self-Hosted (worker-only), Serval-Managed in Your AWS (operator appliance, single-tenant), Self-Managed on Your Kubernetes (public Helm chart)
  • Integrations: 65–70 OOTB (per Apr 2026 sales demo; prior public framing was “60+”). Slack, Teams, Okta, Google Workspace, Azure AD, GitHub, Jira, Notion, Confluence, HubSpot, ServiceNow, Freshservice, Zendesk, Twilio, Jamf, Kandji, Fleet, AWS, Linear, Ramp, Workday, etc. + Custom App primitive (name + shell + logo + OAuth or API key).
  • Pricing (per Apr 2026 demo): ~$30,000/yr minimum spend; per-user; single license type; no implementation fee; no professional-services fee. No SKU laddering between ITSM / workflow builder / access management. Public pricing page still absent.
  • Founded: 2024
  • HQ: San Francisco
  • Parent corp entity: General Intelligence (DBA Serval) ^[inferred]

People

  • Jake Stauch — CEO, co-founder. Background: ex-Verkada.
  • Alex McLeod — Co-founder. Background: ex-Verkada.
  • Kaz Hishida — joined late 2025, per Serval’s “Why I Joined Serval” post.

Funding

RoundDateAmountLeadValuation
Seed2024$8.7M ^[inferred]First Round + General Catalystn/d
Series AAug 2025 (announced Oct 21, 2025)52M total)Redpoint Ventures$232M
Series BDec 11, 2025127M total)Sequoia Capital$1B

Other investors: Meritech, Bessemer, Box Group, Chemistry, Strike Capital, Sunflower, Operator Partners, Evantic, Sound Ventures, Radical Ventures. Angels: Sabrina Hahn, Alex Clayton, Colin Zima.

Velocity post–Series A: revenue +500% and headcount >3x in 90 days (Series B post claim).

Customers

Named (combined public + Apr 2026 sales-demo intel — see demo extract):

  • Public (blog/press): Perplexity, Clay, Verkada, Mercor, Cribl, Together AI, BILT.
  • Disclosed in demo (new): General Motors, Fox, Notion, Brex, LangChain.

Multiple customers reportedly fully replaced their incumbent ITSM with Serval per Series B post.

Material delta: General Motors is using Serval for onboarding at scale — first publicly-disclosed Fortune-50 traditional-enterprise deployment. Resolves a long-standing open question on whether Serval’s playbook generalizes outside AI-forward tech companies.

Standout metrics:

  • Together AI — 95% of just-in-time infrastructure access requests automated.
  • GM — onboarding at scale (specifics not disclosed).

Product

Per the docs surface inventory, the shipped product is wider than press coverage suggests:

Help-desk + ticketing surface

  • Channels: Slack (deep — slash commands, message shortcuts, Team Inbox, 🔒 internal notes, manual escalation buttons), Teams, Email (DKIM-trust-aware, guest user handling), Web Portal, Phone (Twilio), and MCP.
  • Ticket types: Request + Incident (with auto-linking of related incoming tickets, bidirectional sync to ServiceNow/Freshservice).
  • Co-pilot — embedded AI assistant for human agents inside the ticket pane (drafts responses, runs team-only workflows, summarizes threads, enriches with IdP context).

4 named automation tools

  1. Workflows — deterministic TypeScript. Long-running (hours/days). File I/O. Webhook + schedule + event triggers (ticket-created, ticket-updated). Semantic search across the catalog.
  2. Guidance — runbooks/SOPs that shape help-desk-agent behavior; includes an “always-used” flag for tone/compliance/routing rules.
  3. Knowledge Base — Notion, Confluence, Google Drive, M365/SharePoint, ServiceNow, Zendesk, Freshservice, plus custom knowledge-ingestion workflows. Permission-respecting (document-ACL propagation).
  4. Access Management — JIT app/role/resource provisioning via Linked Group / Direct API / Custom Workflow / Manual. Profiles (eligibility) + Policies (rules) + Provisioning (mechanism). Self-positioning: “adjacent to PAM but not a replacement for PAM’s vaulting/session recording.”

Other shipped surfaces

  • Suggestions — AI-generated automation suggestions from ticket patterns (“automate the automation”). Counterpart to Console Assistant.
  • Campaigns — outbound proactive surface: Slack DM broadcast w/ structured workflow response. Counterpart to Console’s Proactive Playbooks. One-time + recurring + local-time-per-recipient.
  • Assets — already shipping (visible as a sidebar section per platform-orientation docs). Earlier read of “telegraphed roadmap” was wrong.
  • Catalog in app.serval.com — Access + All Services (workflows-with-form-inputs surfaced as catalog items). The ServiceNow service-catalog equivalent.
  • CLI — Homebrew-installed (brew install --cask servalhq/serval/serval); OAuth device-flow auth; serval pull/push <team-prefix> produces a team.yaml + workflows/<slug>/{index.ts, workflow.yaml} tree designed for Git workflows; serval access search/request/list/cancel for terminal-driven JIT access.
  • MCP server — public endpoint https://public.api.serval.com/mcp/. End users connect Claude.ai (Custom Connectors), Claude Desktop, or any MCP client to create tickets / check status / list pending access requests via natural conversation. Genuinely novel for the category.
  • Inbox — in-product notification center.
  • Analytics — AI resolved / AI assisted / Unassisted / Resolved-outside-Serval; TTR p25/p50/p75/p95/mean; SLA met/breached; feedback (positive/negative); time- and money-saved estimation; CSV export.

Architecture (corrected vs press)

Two agents per official docs (Product Security):

  • Help-desk agent — handles end-user conversations. Cannot author or modify workflows.
  • Automation agent (the official term — TechCrunch’s “Builder agent” was an informal label) — admin-facing workflow builder; generates TypeScript from NL. No access to end-user ticket context.

Workflows are deterministic by default. “No LLM in the default runtime path” — the docs are explicit. An LLM only appears in a run if the IT admin intentionally adds a step that calls one. Stronger architectural claim than I had earlier.

Integration proxy — credentials live server-side. Workflow code never sees API keys/OAuth tokens. API scoping is fixed at integration setup; no workflow can exceed it regardless of code.

RBAC: 2 org roles (Member, Admin), 5 team roles (Agent, Viewer, Contributor, Builder, Manager), per-team capabilities toggle.

Self-hosting: Three real delivery models — Hybrid (worker-only Linux host, smallest footprint), Serval-Managed in Your AWS (single-tenant operator appliance), Self-Managed on Your Kubernetes (public Helm chart). Day-2 ops docs are gated to customers.

Roadmap still implied by Series B: deeper expansion into HR / Finance / Legal / Security / Engineering automation surfaces (with assets now confirmed live, not roadmap).

Onboarding speed (per demo): systems connected in a 60-minute implementation call; some customers running onboarding workflows in production the next day; smaller organizations running within a few days. Serval is anchoring a faster TTV story than Console’s “demo-to-production-in-3-weeks” claim.

Journeys (newer release, demoed Apr 2026): three views (employee / manager / system) for cross-system onboarding/offboarding. Tasks span security questionnaires, hardware selection, MDM enrollment, 1-on-1 scheduling, group-access grants, Okta account creation, and Google Workspace provisioning. GM uses this surface at scale.

Why It Matters for initlabs

The strongest, best-funded, most strategically positioned direct competitor surfaced so far. Beyond Console in funding, valuation, investor signal, and breadth of stated ambition. Sequoia’s explicit ServiceNow-parallel framing means Serval is being underwritten as a platform replacement, not a feature company.

Competitor Profile

Compare Pages They Maintain

Serval does not appear to publish head-to-head pages publicly (as of research). They position by category framing (“AI-native vs AI-bolted-on”) rather than direct competitive comparison.

Architecture: How They Win

Slack/Teams/Email/Phone/Web/MCP -> Help-Desk Agent -> tool catalog -> deterministic TypeScript run
                                          ^                   ^
                                          |                   | tools published by
                                          |             Automation Agent -> TypeScript + tests + workflow.yaml
                                          |
                                  Ticket = audit record
 
Integration proxy: credentials server-side; API scope fixed at integration setup; workflow code never sees keys

Key primitives:

  • Two-agent architecture (official terms: Help-desk agent + Automation agent) — strict separation. Bounds non-determinism to tool selection, not tool behavior.
  • TypeScript + tests + YAML metadata as the canonical tool format. CLI pulls/pushes a Git-friendly tree.
  • Integration proxy — server-side credential injection; workflows cannot widen API scope beyond what was approved at integration setup.
  • Vibe coding — natural language to working tool, demoed at JNUC as under 60s from screenshot to working Jamf API tool.
  • Access management as first-class product, not integration.
  • MCP as a first-class user-facing surface — Serval-as-tools-inside-the-user’s-AI, not just user-talks-to-Serval-AI.
  • Three real self-hosting models — Hybrid worker, Serval-Managed AWS, Self-Managed K8s with public Helm chart.
  • External ticketing sync — bidirectional with ServiceNow/Freshservice.
  • Ticket-as-audit-record — same philosophical stance as Console.

Positioning Narrative

  1. “AI-native vs AI-bolted-on” — replatform, do not augment. Direct shot at ServiceNow’s AI add-ons.
  2. “Faster to automate forever than to do manually once” — the one mantra Stauch repeats in every interview.
  3. “Vibe coding for IT” — they want to own this category term.
  4. System of record, not feature — Series B reframing, deliberately invokes the ServiceNow parallel.
  5. Outcome automation > step automation — same line as Console; differs in execution (Serval surfaces code, Console hides it).

How They Position Against Each Competitor Tier

TierServal’s Line
Incumbents (ServiceNow, Freshservice, JSM)“Re-platform, don’t bolt AI on. They never reached the bar where automation is faster than manual.”
AI peers (Moveworks, Aisera, Fixify, Console)Largely silent — Serval competes on velocity and funding signal, not direct call-out.
Workflow builders (Zapier, Workato, Tines)Implicit — “we build the workflow from natural language; you don’t drag boxes.”
JIT-access tools (Opal, Tori, Lumos)“Absorb access management into ITSM.”
Identity (Okta, Azure AD)Complementary — Serval consumes IdP context, doesn’t replace IdP.

ICP

  • Size: Mid-market to enterprise. Customer logos skew toward AI-forward, well-funded, often security-paranoid orgs.
  • Stack assumption: Slack-native, Okta-anchored IdP, MDM (Jamf-strong narrative), engineering-forward IT teams.
  • Buyer: Head of IT / VP IT, often with CISO and Engineering leadership co-influence due to security narrative.
  • Geography: US / SF Bay Area dominant.
  • Industry skew: AI-forward tech companies, secondarily security-conscious enterprises.

Strategic Implications for initlabs

  • Pricing is now partially exposed: about $30k/yr minimum, per-user, single license, and no SKU laddering or implementation/professional-services fees.
  • GM in production for onboarding at scale means Serval’s playbook generalizes beyond AI-forward tech companies.
  • Time-to-value benchmark to beat: 60-minute implementation call to systems connected; next-day production for onboarding workflows for some customers.
  • Highest direct threat in the wedge — beyond Console in funding, valuation, investor signal, and breadth of stated ambition.
  • Sequoia is underwriting the “next ServiceNow” thesis publicly.
  • Defensible primitives — vibe coding, two-agent architecture with deterministic runtime, integration-proxy credential isolation, JIT-access depth, MCP server, CLI/Git workflows, three self-hosting models — are now table stakes for serious players. ^[inferred]
  • External-ticketing sync is strategically important because Serval can co-exist with ServiceNow/Freshservice rather than requiring rip-and-replace.
  • MCP as a user-facing surface is a real differentiator vs Console and worth tracking carefully.

Open Questions

  • True ARR and gross margin.
  • Headcount absolute number.
  • Customer-replacement specifics — which incumbent ITSMs were fully replaced, at what customer sizes?
  • How does Automation-agent code reconcile with version control across many tools as the catalog grows?
  • How permissioned is the Automation agent itself — who in the org can vibe-code which automations?
  • Insights vs Analytics — demo nav shows both as separate items; scope of Insights not detailed.
  • International / non-US-Bay traction.
  • Does Sequoia-backed velocity continue post-honeymoon, or does ServiceNow / a M&A move arrive?

Deeper Reading

Graph View

Backlinks