Platform Overview
Execution modes
Mimic supports two complementary execution modes:
RPA: deterministic browser steps generated from recordings and script edits.OpenClaw: agentic execution with chat, search, and tool orchestration.
Both modes share identity, billing, telemetry, and organization scoping.
Core product surfaces
- Studio: author and refine workflows.
- Agents: provision and manage runtime workers.
- Jobs: launch, monitor, and export executions.
- Workflow Packs: package runtime bundles, publish versions, and pin installations.
- Schedules: define cron-based automation with retries and runtime limits.
- Pilot: track enterprise validation, ROI, and release-readiness metrics.
- Feedback Queue: capture operator issues/requests and drive triage automation.
- Settings: billing, credentials, tenant controls, and VM idle policy.
Multi-tenant boundaries
- Every user belongs to an organization.
- Every agent, workflow, and job is organization-scoped.
- API handlers enforce ownership checks before reads or writes.
- Role gates narrow surface area.
dashboard_onlyusers are limited to their configured org dashboard route family and cannot call/api/*.
Data flow summary
- A workflow is authored in Studio.
- A job launch request is submitted.
- An agent executes steps and emits status callbacks.
- Job results and telemetry are stored for reporting and billing.
- Pilot dashboards aggregate outcomes into SLA and ROI views.
Reliability model
- Deterministic step checks for RPA nodes
- Retry-safe APIs and webhook consumers
- Zod validation on API payloads
- Structured status transitions for job lifecycle tracking
- VM safeguards: JIT wake on dispatch, idle auto-stop, stale-run timeout watchdog
- Windows session safeguard: guarded
MimicSessionWatchdogskips forced logoff whileMimicRun_*automation is active - DB resilience hardening: targeted high-traffic indexes and larger connection pool sizing for concurrent agent workloads
When to use each mode
- Prefer
rpafor strict, repetitive portal sequences. - Prefer
openclawwhen tasks require branching, reasoning, or long-form interaction. - Many teams run both: RPA for high-volume deterministic tasks and OpenClaw for exception handling.