Key facts
| Agent stack | Chat completions, function calling and embeddings are live |
| Use cases | Disruption option summaries, passenger comms drafts, maintenance log triage, policy Q&A |
| Human control | Operational and safety decisions stay with authorised staff |
| Tools | Expose reservation, policy and log systems as read-only tools |
| Models | 30+ models; small aliases for triage, stronger ones for multi-source synthesis |
| Deployment | Plugsky cloud, your VPC, on-prem and air-gapped options |
| Pricing model | Flat monthly self-serve plans with unlimited fair-use usage |
| Roadmap | Assistants and batch endpoints are coming soon |
TL;DR
- Use agents for preparation and communication, not operational decisions.
- Expose read-only tools for policies, schedules and maintenance records.
- Require citations so staff can verify every recommendation.
- Build escalation paths for irregular operations where rules conflict.
- Measure agent usefulness by staff time saved, not calls answered.
How it works, step by step
- Pick a bounded workflow such as maintenance write-up triage or policy Q&A.
- Expose read-only tools for manuals, policies and records, with no write access.
- Ground answers in retrieved documents and require a citation for every claim.
- Draft outputs for a dispatcher, engineer or agent to review and act on.
- Add escalation rules for conflicting policy, safety items and unusual cases.
- Log all tool calls and model versions, then measure review time and error rate.
Try it yourself
Open the multi-agent workflow generator →
Where airline agents add value
Airlines have procedure-heavy knowledge work and time-critical communication:
- Disruption support: assemble rebooking options and policy constraints for a human to choose from.
- Passenger communication: draft clear, compliant updates during irregular operations.
- Maintenance triage: classify write-ups by system and severity, and surface the relevant manual section.
- Policy Q&A: answer handling, baggage and fare questions with citations to the current rule.
Every one of these informs a person who remains accountable.
Architecture with hard human control
Keep the agent on the preparation side of the decision line:
- Retriever: manuals, policy, schedules and sanitised historical cases.
- Read-only tools: lookups that cannot change a booking, a work order or a flight.
- Planner: decomposes a request into lookups and a structured recommendation.
- Composer: produces a draft with citations and explicit assumptions.
- Human decision point: a dispatcher, engineer or agent authorises the action.
Any write action goes through your existing systems and approval process, never through the model.
Safety, escalation and audit
Define escalation before launch: conflicting policies, safety-related items and anything outside the indexed corpus should route to a person immediately. Keep prompts and model versions pinned so behaviour is reproducible, and log every tool call with its inputs and outputs. Where a regulator or internal policy requires isolation, run the workload in a VPC, on-prem or air-gapped environment using the same compatible API. Treat passenger data carefully too: send references where possible and keep unnecessary personal details out of prompts.
Pilot and evaluation
Prototype with plugsky-micro or plugsky-lite on one team's knowledge base, then use the 14-day full-access trial to test a stronger model on multi-source synthesis. Track triage accuracy, citation support rate, staff review time and escalation rate. Self-serve plans are flat monthly with fair-use usage, so peak disruption periods do not create billing surprises. Roll out team by team with versioned prompts, and re-run the evaluation set when manuals, policies or models change.
Honest comparison
| Capability | Plugsky | Typical agent framework | Building in-house |
|---|---|---|---|
| Tool calling | Live function calling over read-only internal APIs | Framework-specific adapters | You build and maintain |
| Retrieval | plugsky-embed plus RAG over manuals and policy | Add-on or external store | You integrate everything |
| Human control | You enforce approval gates in code | Varies | You design controls |
| Deployment | Cloud, VPC, on-prem and air-gapped | Usually cloud-only | You operate GPUs and serving |
| Pricing | Flat monthly self-serve, fair-use usage | Varies by framework and model | GPU plus operations cost |
Frequently asked questions
Can the agent make operational decisions?
No. Keep it to retrieval, drafting and recommendations; authorised staff remain accountable for operational and safety decisions.
Which endpoints do airline agents use?
Chat completions, function calling, JSON mode and embeddings are live. Assistants and batch endpoints are coming soon.
How do we handle irregular operations?
Predefine escalation rules for conflicting policy, safety items and out-of-corpus requests, and route them to a human immediately.
Is there a free plan?
Yes. The free plan includes plugsky-micro and plugsky-lite with no credit card, enough to prototype knowledge retrieval and triage.
How does pricing work?
Self-serve plans are flat monthly with unlimited fair-use usage. See the live pricing page for current plans and enterprise options.
Can we run it in our own environment?
Yes. VPC, on-prem and air-gapped deployments are available with the same compatible API and application code.
How should we measure it?
Track triage accuracy, citation support, staff review time and escalation rate. Time saved per case is the number that matters.