Your software holds the work. Help us prove the handoffs.
Arnexyia is a vendor-neutral operations layer for specialty subcontractors and growing service businesses. Its founding pilot is designed to test repeated handoffs between the tools you already use. Current engineering validation is limited to synthetic shadow workflows and a controlled Gmail read-only test path. Calendar, CRM, field-service and document integrations remain future pilot work, selected only when a qualified workflow requires them.
This is not another system of record and it is not a pitch to replace your staff. Your existing tools stay in place. Discovery identifies where coordination breaks down and what evidence a useful pilot would need. Risk tiers, deterministic checks, audit events and a shadow-testing period control what the system may do across those systems.
- Default execution mode
- Shadow
- Capabilities under policy
- 17
- Synthetic test fixtures
- 51
Proposals only, no side effects
Each green, yellow or red per client
Replayed with no client data
What you are actually buying
Fewer dropped handoffs across your existing stack — measured against your own baseline.
We do not publish savings figures or error rates, because we have not earned them yet. During onboarding we record where work moves between systems, your current response times, intake completeness and pipeline hygiene, then report the delta from that baseline. Every number in a report traces back to a record in your own workspace.
You grant the permissions
Each integration is connected by you, with the narrowest scope that makes the work possible. Read-only first wherever the work allows it.
You define the authority limits
The authority matrix is a per-client document: which actions are autonomous, which need a second check, which always need you, and the dollar thresholds that apply.
You can stop everything
A global kill switch halts autonomous execution across all clients immediately. Per-capability live mode can be withdrawn at any time.
Operational coverage
Current scope and potential pilot coverage
Inbox & follow-up
The current Gmail test path can normalize one authorized message and evaluate a shadow proposal. It does not send replies; the controlled connection and read test remain pending.
Scheduling
Potential pilot coverage: visit-window and conflict recommendations after a supported calendar connection is selected, authorized and verified. Calendar access is not currently connected.
CRM hygiene
Potential pilot coverage: identify missing records and stalled stages from client-authorized evidence. A CRM or field-service connector will be chosen from a qualified pilot’s actual stack.
Bid & intake coordination
Potential pilot coverage: identify bid deadlines, missing documents and coordination gaps. Document ingestion and quote preparation require separate implementation and validation.
Client communication
Potential pilot coverage: proposed acknowledgements and status updates for human review. No customer messages are sent automatically; complaints and safety matters remain human-led.
Reporting
The console supports recorded outcomes, baselines and intervention tracking. Customer results remain unproven; reports must distinguish synthetic records from verified observations.
Exception & handoff monitoring
The shadow engine records policy decisions and proposed actions. Cross-system completion and stalled-handoff monitoring are pilot hypotheses to validate after the relevant integrations exist.
Nothing outside the matrix
If an action is not listed in your authority matrix, the system will not attempt it. Unknown action types are blocked, logged and raised as an exception for a human.
Control model
High autonomy is only safe with hard limits around it
Shadow mode is the default
Current workflows produce shadow proposals and expected outcomes. No live executor exists. Sending, booking, charging, publishing or deletion cannot be enabled by a customer setting.
Deterministic guardrails run first
Tenant match, recipient and domain validation, duplicate prevention, amount thresholds, required fields, date validation, opt-out status, per-hour rate limits, unknown-action blocking and prompt-injection screening are all implemented as deterministic checks that run before any model judgement.
Risk-tiered authority
Each capability is green, yellow or red per client. Banking changes, refunds, contract acceptance, legal admissions, regulated representations and destructive deletion stay red permanently.
Everything is on the record
Each action writes an audit event with actor, tenant, inputs, decision, the policy rule applied, confidence, intended side effect, outcome and an idempotency key.
Test gates before live mode
Tenant isolation, duplicate prevention, opt-out honouring and high-risk blocking must each pass at 100%, with at least 95% overall, on synthetic fixtures. That only unlocks real-agent shadow testing; real-data shadow, legal review, durable runtime, observability and written per-capability approval all come before any limited live use.
Founding pilot intake is open
Founding clients begin with discovery, baseline planning and written authority limits. Controlled read-only client data is connected only after the applicable release, processor and legal gates clear. You see proposed actions and expected outcomes before any capability is considered for live execution.