Pack credits
dpanel.ai Pack
Pack uses prepaid credits, pay-for-what-you-use metering, and a pre-flight hold that blocks calls the balance cannot cover.
Guardrails
Guardrails
Workspace-default policy applies across instances, with per-instance overrides for projects that need stricter usage or safety rules.
Guardrails
Budget Policies
Set spending limits and rate caps to control usage costs with daily and monthly caps plus request and token rate limits.
Guardrails
Model & Provider Access
Control which models and providers are available before any Pack or BYOK model call can run.
Guardrails
Prompt Injection
Detect and respond to prompt injection attempts, jailbreaks, and instruction-override attempts before the model runs.
Guardrails
Sensitive Info Detection
Identify and handle PII, credentials, and personal data by flagging, redacting, or blocking requests.
Human coworkers
Humans + AI, one team
Invite human coworkers, assign owner/admin/manager/member/viewer roles, and keep accountability visible beside AI workers.
Strict policy
USER.md rules can be locked
Owner/admin policy rights can make a grounding non-overridable for the worker. Others can see it without silently changing it.
Per-profile channels
One channel binding per profile
Each AI worker or human profile can carry its own WhatsApp, Telegram, Slack, or email channel context for cleaner handoff.
Knowledgebase
DPanel Pack vs BYOK
DPanel Pack is the optional managed model-usage path with prepaid credits and metered calls. BYOK stays the default path and DPanel never charges BYOK customers for model usage.
How Pack Billing Works: Credits, Holds, and MarginPack billing uses prepaid credits, pre-flight holds, metered model calls, and transparent margin so customers can control managed AI-team spend.
Use Your Provider Key on a DPanel InstanceApply a connected BYOK provider key to one of your own DPanel instances with explicit consent. The key goes into your instance only and DPanel never charges BYOK model usage.
Where Can My DPanel Instances Run?DPanel is the control plane, not the hosting company. Instances can run locally, on an existing remote host, or on supported cloud and partner targets as they become available.
Guardrails OverviewDPanel Guardrails apply usage and safety policy server-side before every model call, with a workspace default and per-instance override model.
Budget PoliciesBudget Policies set monthly and daily spend caps plus request- and token-per-minute rate caps for DPanel model calls.
Model & Provider AccessModel & Provider Access policies allow or deny the model and provider paths each DPanel instance may use.
Prompt Injection GuardrailsPrompt Injection guardrails detect jailbreak and instruction-override attempts before model calls run.
Sensitive Info DetectionSensitive Info Detection identifies PII, credentials, and personal data in prompts so DPanel can flag, redact, or block the request.
課題
- 課題
- 承認
成果
- 危険な操作前の正式な承認ゲート
- 引き継ぎ出力
- ツール、タスク、チャネル、実行先を横断
AIチームメンバー
Vault guideRuntime operatorUsage monitor
ワークフロー
連携
OpenAIAnthropicGeminiOpenRouterProvider vault
承認ポイント
- Before using sensitive keys
- Before changing default provider
引き継ぎ
Provider selected, model class, usage state, and any required approval.
FAQ
What is BYOK AIチームデプロイ?
既存プロバイダーや直接のコスト管理が必要なチーム向け。
Who is this dpanel.ai page for?
既存プロバイダーや直接のコスト管理が必要なチーム向け。
What does the AI team actually do?
It helps with 危険な操作前の正式な承認ゲート, 引き継ぎ出力, ツール、タスク、チャネル、実行先を横断, while dpanel.ai keeps tasks, approvals, and handoff visible.
Where do approvals happen?
Approval moments include before using sensitive keys, before changing default provider.
What does handoff look like?
Provider selected, model class, usage state, and any required approval.