FAQ

Positioning

Q: combo agent vs ChatGPT / Claude? A: combo agent is not a vanilla chat assistant—it is an enterprise production platform: skills, toolchain integrations, teamwork controls, auditable orchestration atop foundation models.

Q: Relationship between xIPnex and combo agent? A: combo agent is the platform; xIPnex is the patent vertical. Automotive and office suites follow the same pattern.

Q: Which LLM backs the stack? Can we swap models? A: Model-agnostic adapters cover DeepSeek, Qwen, GLM, Claude, GPT, etc. Private deployments may run wholly offline stacks (vLLM / Ollama).

Q: APIs for extensions? A: REST + webhooks + CLI + Python SDK—see API reference.

Data & compliance

Q: Is my data used to train vendor models? A: No for SaaS operational data; on-prem workloads never leave customer networks.

Q: GDPR / PIPL / classified programs? A:

• Classified-ready private builds
• GDPR/PIPL: tenant isolation, residency options, auditing, deletion rights
• Domestic stack kits for localized hardware/software

Q: Sensitive payloads reaching LLMs? A: Mitigations include on-prem inference, redaction Hooks before model calls, and KB-only workflows without cloud inference.

Q: Cross-border data rules? A: Tenant-level residency; EU workloads can remain on EU nodes end-to-end.

Usage

Q: What is a Skill? A: Encapsulated professional capability (patent authoring, reviews, ALM sync). Details: Skills concept, catalog.

Q: Reuse artifacts across sessions? A: /graft—see Graft skill.

Q: Auto-review every PR? A: Event triggers plus code-hosting integrations (Events, Hosting).

Q: Cron-style jobs? A: Platform Cron or tool-assisted scheduling—Cron concepts.

Q: Restrict agents to enterprise data only? A: Combine Hook tool allow-lists, prompt fences, and KB-scoped retrieval.

Deployment

Q: Private hosting? A: Yes—Plans.

Q: Hardware sizing? A: Minimal 8c32 GB + ≥100 GB SSD; prod typically 16c64 GB + ≥500 GB SSD + GPUs if local LLMs — Installation.

Q: Kubernetes? A: Helm chart documented in Installation.

Q: Air-gap installs? A: Offline image bundles ship with deliveries.

Q: Upgrade downtime? A: Rolling upgrades; DB migrations flagged ahead—budget ~10 min worst case—see Monitoring.

Security

Q: Supported login schemes? A: OIDC, SAML 2.0, LDAP / AD, embedded IdPs.

Q: Prompt leakage risks? A: Hosted proxy plus Hook redactions; audited model calls.

Q: Departing employee cleanup? A: Disable accounts, optional session ownership transfers, tenant purge flows.

Q: Audit coverage? A: Authentication, ACL changes, tooling, KB access, external calls—Security.

Pricing

Q: Billing models? A:

• SaaS: token + session tiers
• On-prem: annual enterprise license by scale/seats
• Professional services: time & materials

Q: Monthly estimate? A: Pilot calculators combine expected sessions × average tokens × model rates—ask sales for spreadsheets.

Q: Spending guardrails? A: Tenant / user / agent quotas plus alerts.

Integrations

Q: DOORS / Jira / Polarion? A: Covered under DOORS integrations.

Q: Use both Feishu and WeCom? A: Yes—dual channels bind to shared agents/KBs.

Q: Custom enterprise systems? A: Bidirectional webhooks (Webhook guide) or bespoke Skills over internal REST.

Q: GitLab private + GitHub cloud simultaneously? A: Separate credentials plus routing rules.

Incident response

Q: Agents hang without answering? A: Inspect Ledger—usually stalled LLM/tool dependencies.

Q: Erratic outputs? A: Upgrade model tier, tighten Skills, enforce Hook validations / confidence gates, escalate sensitive steps for human confirmation.

Q: Permission denied? A:

• IM scopes / visibility windows
• Gerrit labels
• DOORS Project Area ACLs — see respective integration pages

Still stuck?

→ Contact us