Fig. 4.1 
Multi-tenant control plane
One layer, many teams, zero data bleed
Tenants, users, sessions, keys, memory and files stay separated — so internal teams and clients share the same operating layer without ever sharing data.
dewee runs agents across every provider and channel — with tenant isolation, real permissions, cost evidence, and audit trails — so teams move fast without losing control.
1
Tenant isolation
Each workspace keeps its own users, keys, memory and files. No shared data.
2
Permission nodes
Scoped grants and approval gates in front of every tool an agent can reach.
3
Provider routing
Bring your own keys across any model — local or enterprise endpoints.
4
Audit & cost trail
Every action recorded with the model that ran it and what it cost.
- 1
Tenant isolation
Each workspace keeps its own users, keys, memory and files. No shared data.
- 2
Permission nodes
Scoped grants and approval gates in front of every tool an agent can reach.
- 3
Provider routing
Bring your own keys across any model — local or enterprise endpoints.
- 4
Audit & cost trail
Every action recorded with the model that ran it and what it cost.
Routes to the providers & channels your teams already run
20+
Provider routes & OpenAI-compatible endpoints
7
Chat-native channel families in the gateway
5
Defense layers before tool execution
D1·R2·MCP
Cloudflare data, media & agent inspection
Fig. 2 — The gap
A chat window can answer.
It can't run a company.
The moment agents act on real systems — money, data, customers — "just a chatbot" turns into three problems no prompt can fix.
- 01
No isolation
One workspace, one key, everyone's data mixed together. Fine for a demo, unacceptable for clients.
- 02
No permission
An agent that can read a doc can usually also delete it, post it, or pay for it. Nobody approved that.
- 03
No evidence
When spend spikes or something ships wrong, there's no trail of who acted, with which model, and why.
Fig. 3 — How it works
Three moves from prompt to accountable action.
- 01
Connect
Bring your provider keys and plug in the channels your teams already use — OpenAI, Anthropic, Gemini, local models, Slack, Discord, Telegram and more.
- 02
Govern
Carve out tenants, set scoped permissions and approval gates, and put budgets and policy in front of every tool an agent can reach.
- 03
Act with evidence
Agents and people work the same queue. Every action lands with an audit trail, a cost record, and the model that ran it.
Fig. 4 — The platform
Four things a chat window will never give you.
Fig. 4.2 
Security-first
Five checks before any tool runs
Policy, injection detection, credential scrubbing, scoped grants and sandboxing sit between intent and execution.
Fig. 4.3 
Human + agent teamwork
Lead agents delegate, people approve
Agents create tasks, hand off work and collect results on shared boards — with humans kept in the decision path.
Fig. 4.4 
Provider freedom
Your keys, your models, your bill
Route across OpenAI-compatible, Anthropic, Gemini, Groq, DeepSeek, local or enterprise endpoints — no LLM credits bought from dewee.
Knowledge vault
Documents, memory and knowledge-graph retrieval, tenant-scoped.
Cost observability
Per-run token, cost and fallback evidence your finance team can read.
Channels & webhooks
Slack, Discord, Telegram, WhatsApp, Zalo, WebSocket and HTTP APIs.
Fig. 5 — Security-first by design
Permissions, audit trails, and review before any tool runs.
dewee puts five defense layers between an agent's intent and execution — policy checks, credential scrubbing, scoped permissions, approval gates, and audit evidence — so operators stay in control of every action.
Fig. 6 — Enterprise comparison
Where dewee sits beside GoClaw, OpenClaw, and Hermes.
A positioning table, not a benchmark. OpenClaw and Hermes are strong self-hosted/autonomous projects; dewee is shaped around enterprise governance.
| Capability | dewee | GoClaw | OpenClaw | Hermes |
|---|---|---|---|---|
| Enterprise tenancy and RBAC | Advanced tenant isolation, permissions, and admin control-plane | Core runtime maintained for stability | Personal/self-hosted gateway focus | Personal autonomous-agent focus |
| Security-first agent operations | Policy, audit trails, credential scrubbing, and deployment review | Runtime hardening and production security layers | Depends on local/self-hosted setup | Depends on operator deployment and tool permissions |
| Human and agent teamwork | Lead agents, member agents, approvals, and shared work evidence | Agent teams and task boards continue in maintenance path | Single gateway runtime model in public docs | Autonomous agent with persistent memory and skills |
| Provider strategy | BYO API keys and enterprise provider routing | Broad provider support | Bring your own models/providers | Configurable providers and toolsets |
| Knowledge layer | Vault, documents, memory, and AgentBrain deep integration path | Knowledge features stay stable | Workspace/session-store oriented | Persistent memory and self-improving skills |
| Commercial support | Enterprise support, onboarding, and self-hosted option | Community/open runtime maintenance | Open-source community model | Open-source ecosystem |
A chatbot answers. An operating layer is accountable — for who acted, with which model, under what permission, at what cost.
The dewee thesis
Fig. 8 — Created by NextLevelBuilder.io
Built by operators who ship AI products, education, and infrastructure.
NLB builds from live product pressure — customer workflows, agent tooling, deployment automation, and knowledge-heavy systems held to enterprise expectations.
Duy Nguyen
NLB founder, product and engineering
Builds agent tooling, product automation, and AI-native engineering systems.
Cuong Vo
NLB operator and growth advisor
Brings product launch, pre-order, and automation operating experience.
Viet Tran
Architecture advisor
Supports high-load systems thinking, education, and enterprise architecture.
Fig. 9 — Vision and roadmap
From agent gateway to governed operating layer.
GoClaw stays the stability path. dewee is the enterprise branch where permissions, tenancy, and knowledge features move faster.
- Now
dewee private enterprise platform, marketing site, and hosted control-plane boundary.
- July 2026
More advanced permissions, admin analytics, tenant onboarding, and agent website automation.
- August 2026
emree coming soon: pro-user AI coworker across providers, browser, desktop, and dewee control.
- Later
Optional bundled LLM subscription research, deeper enterprise deployment options.
Fig. 10 — FAQ
Questions customers ask before choosing dewee.
This FAQ summarizes deployment boundaries, refund policy, provider keys, On-Premises scope, and privacy before teams move into app.dewee.sh.
- How is SaaS different from On-Premises?
- SaaS runs in dewee-managed shared hosting. Dedicated TOSE runtime gives a stronger isolated runtime boundary. On-Premises is installed on customer infrastructure.
- What is the refund policy?
- SaaS subscriptions can request a refund within 14 days of purchase, subject to abuse, heavy usage, completed setup work, payment processor, and legal exclusions.
- How do teams evaluate before buying?
- Teams can review pricing, read the policy pages, and book a demo with NextLevelBuilder.io before subscribing or requesting an On-Premises quote.
- Does dewee include LLM credits?
- Not yet. Customers bring their own AI provider API keys and control provider spend inside app.dewee.sh.
- What is included in the On-Premises estimate?
- The public ~$5K++ estimate includes installation on customer server or computer, around 5 custom workflows, annual license, maintenance, and updates for 1 year.
- What about data, security, and privacy?
- dewee focuses on tenant isolation, permission controls, audit evidence, provider-key boundaries, and public Terms, Privacy, and Refund policy pages.
Fig. 11 — Related products
The NLB ecosystem around dewee.
dewee sits next to knowledge, agent-workflow, and deployment products instead of becoming an isolated chatbot island.
AgentBrain.sh
Enterprise Knowledge Base and Knowledge Graph layer that can integrate deeply with dewee for retrieval, memory, evidence, and organizational context.
GoClaw
The open platform maintained for fixes, stability, and community runtime workflows.
ClaudeKit
Premium skills and workflows for Claude Code.
AgentWiki
Knowledge base, documentation, and publishing workflows for agent-native teams.
UI UX Pro Max
Design intelligence for professional UI/UX systems.
TOSE
Fast production deploys with Kubernetes, SSL, and rollback.
Fig. 12 — Start with dewee
