Self-HostedComing soon
Your infrastructure. Your data. Your AI.
Run the full platform — chat, agents, knowledge base, everything but the marketplace — inside your own environment. Nothing leaves your network unless you decide it does.
Self-hosted deployments are coming soon — early conversations start now, and they shape the rollout order.
3
Ways to run it
Compose, Kubernetes, your Azure tenant
0
External AI calls
With the local Ollama stack
9
Model providers included
Including local inference
Sorted by constraint, not by industry
If your data residency policy, regulatory obligations, or security review would block a shared multi-tenant cloud, self-hosted removes that question entirely. Your data stays in your PostgreSQL database, your files in your blob storage, and your provider API keys encrypted at rest — inside a network boundary you control. With the Ollama stack, even inference never leaves the building.
What the self-hosted edition includes
- Multi-model chat across all 9 providers — including fully local inference with Ollama
- Custom agents with versioning, publish, and rollback
- Knowledge bases with cited, grounded answers
- Memory, projects, and the file library
- Slack, Teams, and embeddable webchat deployment
- MCP servers and integrations
- SSO/SAML/OIDC, SCIM deprovisioning, RBAC, and the always-on audit trail
Three ways to run it
Docker Compose
Three stacks, from a minimal footprint to a full one — including an Ollama stack where the entire platform runs on a local LLM with zero external AI calls.
Kubernetes
Manifests with kustomize overlays for every service, including an isolated sandbox workload template for code execution.
Your own Azure tenant
Infrastructure-as-code deployment into your subscription — your PostgreSQL, your blob storage, your Key Vault, your network boundary.
Licensing is per named user — you license the software, not the tokens, because inference runs on your own keys. Deployment specifics are covered in the self-hosting guide.
FAQ
Self-hosting questions
Short answers, nothing hiding in a footnote.
Self-hosted deployments are coming soon, and they're licensed and provisioned through a conversation rather than a public download. Deployment artifacts, environment scripts, and migration tooling come with the engagement. Early conversations start now, and they shape the rollout order.
Yes. The Docker Compose Ollama stack runs the same platform plus a local Ollama container, with zero external AI calls — the whole thing answering from a local model, with nothing leaving the machine.
Per named user. You license the software, not the tokens, because inference runs on your own provider keys — or entirely locally with Ollama. There's no per-token markup to anyone. It's genuinely custom to your infrastructure, so we don't publish a number.
Because the self-hosted build physically excludes the billing and marketplace plane. That isn't an asterisk — it's exactly what lets the deployment keep serving chat, agents, and knowledge even when it can't reach us at all.
Your PostgreSQL database, your blob storage, and your provider API keys encrypted at rest — inside a network boundary you control. GDPR erasure tooling and the audit trail run inside that boundary too.
Yes — SSO over SAML/OIDC with DNS-verified domains, and SCIM for deprovisioning. Auto-provisioning is on the roadmap rather than shipped, and we'd rather say so than imply otherwise.
Yes. Agents move between cloud and self-hosted as portable export packages, so an agent built in SaaS lands in a self-hosted deployment with its definition intact. Upgrades are versioned releases with migration scripts.
Talk to us about your environment
Self-hosted deployments start with a real conversation — your constraints, your infrastructure, your identity provider — not a checkout page.