The engine inside ServiceAI

OpenClaw does the operating.

An autonomous agent runtime built for service operations. Triggers fire, the model reasons, tools execute, work gets done — at the bottom of a real workflow, not the top of a chat window.

380ms · voice latency · 47 · built-in tools · 99.95% · uptime
01
The four stages

Every workflow OpenClaw runs has the same shape.

It's not a chatbot. It's an agent that gets handed a job, thinks through it, executes against real tools, and reports back. Same loop every time, whether the trigger is a phone call or a calendar gap.

/ 01 · TRIGGER
Something happens.
A call comes in. A review gets posted. A calendar slot opens up. A tax form is overdue. Triggers can be inbound (caller, email, webhook) or scheduled (daily 9 AM digest).
/ 02 · REASON
The agent thinks.
Pulls customer history, business rules, calendar state, and conversation context. Decides what to say, what to ask for, what tools to call. This is the part most "AI receptionists" don't do.
/ 03 · EXECUTE
It does the work.
Calls real tools against real systems: book the calendar slot, send the SMS, charge the deposit, update the CRM, post the review reply. Not a "draft" — the actual action.
/ 04 · REPORT
You hear back.
Full transcript, audio, what changed where, and what comes next — written to your ops console and surfaced in the morning digest. Errors and escalations route to a human in real time.
02
Why it's different

Three things "AI receptionists" don't do.

Voice answering is table stakes. Where OpenClaw actually earns its place is in the messy bits — the modifications, the escalations, the after-hours edge cases.

01
It writes back to your systems.
Most AI receptionists hand you a transcript and a callback button. OpenClaw books the appointment, charges the deposit, updates the CRM, and sends the confirmation — all before the caller hangs up.
02
It runs on a schedule, not just on a ring.
The high-leverage work is proactive: outreach to lapsed customers when the calendar's soft, chasing missing tax docs, drafting review replies. Triggers can be time, calendar gaps, business events, or webhooks.
03
It hands off cleanly to humans.
When the agent isn't confident — a high-dollar quote, an angry customer, anything outside policy — it escalates with a one-line summary, full transcript, and the suggested next action already drafted. No cold transfers.
03
Real workflows

Four examples, one runtime.

Each row is one OpenClaw workflow running in production today. Same engine, four different jobs — built by configuring triggers, tools, and rules, not by writing code.

INBOUND
TRIGGER
After-hours missed call
REASONING
Identify intent, check calendar, propose two slots
ACTION
Book on Google Calendar · SMS confirmation
Booked
REPUTATION
TRIGGER
1★ Google review posted
REASONING
Find the customer's history, classify the complaint
ACTION
Draft reply · route to GM · send on approval
Replied 12m
OUTBOUND
TRIGGER
Slow Tuesday in 3 days (cal < 60%)
REASONING
Pull lapsed customers · personalize the ask
ACTION
SMS reactivation · book replies into open slots
+11 booked
SCHEDULED
TRIGGER
Tax doc overdue from vendor > 7 days
REASONING
Identify vendor, escalate path, polite reminder tier
ACTION
Email + SMS chase · upload to accounting
$4.2K saved
01
Architecture

Six layers, one runtime.

OpenClaw isn't a wrapper around an LLM. It's a stack: ingress, context resolution, planning, tool execution, observation, persistence. Each layer is independently versioned and rate-limited.

L6 Ingress · Voice / SMS / Email / Webhook Twilio · SendGrid
L5 Session Manager · STT, turn-taking, context Deepgram · proprietary
L4 Planner · intent + plan synthesis Claude · GPT-4o · fallback router
L3 Tool Runtime · function calls + retries 47 built-in · custom REST
L2 Observer · guardrails + escalation triggers Rule engine + classifier
L1 Persistence · transcript, audio, events Postgres · S3 · OpenSearch

Designed for voice latency first.

Voice is the hard constraint. Below ~600ms end-to-end and humans don't notice; above and they hang up. We hit a P50 of 380ms by streaming partial STT into the planner and starting TTS on the first sentence boundary.

The planner runs against multiple models in parallel for high-stakes turns — quotes over $1k, escalations, refunds — and uses the fastest valid response. For routine turns it's pinned to the cheapest model that meets the latency SLO.

Tools execute against real systems with idempotency keys and a 3-tier retry policy. Every external write goes through a dry-run shadow pass before commit when a workflow is in "supervised" mode — turn it off once you trust the agent.

02
Defining a workflow

Declarative YAML, or build it in the GUI.

Workflows are versioned config, not prompts. They define triggers, tools the agent is allowed to call, guardrails, and escalation paths. Anything you can build in the visual editor compiles to this.

workflows/after-hours-booking.yaml v2.4 · LIVE
# When a call rings outside business hours, the agent answers as Mike's Auto,
# qualifies the job, proposes two open slots, and books on confirmation.

workflow: after-hours-booking
version: 2.4

trigger:
  type: inbound_call
  when: outside_business_hours || phones_busy

context:
  load:
    - caller.history          # last 5 visits, lifetime value
    - calendar.next_72h       # bookable slots, holds, capacity
    - business.rules          # pricing, services, "we don't do X"

tools:
  - calendar.propose_slots       # read-only, suggest 2–3
  - calendar.book               # write, requires confirmation
  - sms.send_confirmation
  - crm.create_or_update_contact
  - human.escalate              # last resort

guardrails:
  refuse_if:
    - quote_total > 1500          # escalate to GM
    - sentiment.anger > 0.7
    - caller.history.complaints >= 2
  confirm_before_writing: true
  max_turn_count: 12

handoff:
  to: on_call_human
  with: [transcript, suggested_action, caller.history]
03
Guardrails

Four layers between the model and your customer.

The model is one component. What keeps the runtime safe is everything around it — schemas, allowlists, classifiers, idempotency, and a "panic button" that kicks any conversation to a human in <500ms.

Tool allowlist

Workflows can only call tools they've been granted. Calendar-book can't suddenly become charge-credit-card. Every tool has an explicit schema and the planner can't synthesize arguments outside it.

SCOPE_KEY → SHA256 → SIGNED CALL

Confirmation gates

Any write that touches money, calendar, or customer records is confirmed against business rules before commit. New workflows ship in "supervised" mode — every action is shadowed and reviewed for the first 50 runs.

DRY-RUN → HUMAN APPROVE → COMMIT

Escalation classifier

A separate, smaller model runs on every turn watching for anger, confusion, high-value asks, or anything outside the workflow's competence. P95 escalation-to-human time is 480ms — usually before the customer notices.

SENTIMENT · COMPLEXITY · POLICY

Audit log

Every plan, tool call, retry, and external write is logged with content hashes and the exact model version used. Searchable in the console for 13 months. Customer data isn't used for training and is purged on schedule.

SOC 2 TYPE II · 13-MONTH RETENTION

04
Specs

The numbers behind the runtime.

P50 / P95 measured across the last 30 days of production traffic, all customers, all workflows. Updated nightly in the status dashboard.

VOICE LATENCY
380ms P50 · 720ms P95 · end-to-end from caller speech to agent response
TOOL CALL P50
140ms · 320ms P95 · for built-in tools (calendar, CRM, SMS)
CONCURRENT SESSIONS
Unlimited per account · regional capacity headroom 4×
ESCALATION HANDOFF
<500ms to live human · context bundle delivered in <1s
UPTIME (TRAILING 90)
99.97% · SLA 99.95% on Operator+ · public status at status.serviceai.com
LANGUAGES
29 native-fluent · code-switching mid-call · regional accent handling
INTEGRATIONS
47 built-in · plus any REST API via custom tool definitions
DATA RESIDENCY
US-hosted · AWS us-east-1 primary, us-west-2 failover · EU available on Operations
Build your first workflow

Open the Builder. Ship in an afternoon.

The visual builder ships with 14 starter workflows for service businesses. Fork one, point it at your tools, watch it run in the shadow lane until you trust it.