Your company's privateAI workspace.
Chat, team-scoped memory, and 200+ models on your own servers. Every conversation stays on your infrastructure, every team sees only its own knowledge, and admins see usage, cost, and audit logs across the board.
Particula⌄+ New chat
Search⌘K
Todaypgvector decisionQ3 launch checklistOnboarding email draftYesterdayInfra cost review
CCamila R.Org admin
pgvector decisionParticula Iskra ⌄
What did we decide on pgvector, and why?
Searching team memory…
Consolidated on pgvector in April, over a managed vector DB: one less system to run, and recall held at your scale.Ana flagged the HNSW index parameters early; the team benchmarked before committing.Migration closed 14% under budget. [thread · Apr 18] [thread · Apr 22]
✓ Team knowledge · 3 threads cited · answered on-prem
+Message Lumen…↑
Where should we start, Camila?
Particula · fine-tuned
Particula IskraSoon
Anthropic
Claude 4.5 Sonnet200K
Self-hosted · vLLM
Llama 3.3 70B128K
DeepSeek V3.2128K
Draft the Q3 kickoff brief from our launch checklist
＋ Attach◈ Org knowledge◎ ResearchParticula IskraLlama 3.3 70B⌄↑
Particula+ New chat
pgvector decisionQ3 launch checklistInfra cost reviewWhat did we decide on pgvector, and why?
Consolidated on pgvector in April, over a managed vector DB: one less system to run, and recall held at your scale.
Message Lumen…
⌕vector db choiceBy meaning · not keywords
pgvector decision“…consolidated on pgvector over a managed store…”Apr 18
RAG eval harness“…HNSW recall at 10M rows held above target…”Apr 22
Infra cost review“…migration closed 14% under budget…”May 02
↑↓ Navigate · ↵ OpenSearched on-prem
188/200monthly active users41Kmessages$0/seatflat license
Usage by team
Engineering86%
Product71%
Sales64%
Legal38%
Every query stays in your VPC
Audit log · live
09:41Ana M. · restricted GPT-4o for Legal teamPolicy
09:38SCIM · 4 members provisioned from OktaIdentity
09:31Knowledge base · 12 docs indexed for ProductRAG
Streaming from your cluster
What it does
See it live in a demo →Every model, one workspace200+ models via OpenRouter, or your own Ollama and vLLM instances. Swap models without changing workflows; context length and pricing shown per model.
Team-scoped memoryEvery message is embedded and searchable. New conversations start with relevant team context injected automatically, with attribution: who said it, when, and where.
Admin visibilityUsage and cost per model, team, and user. Conversation browser, audit log, and per-team model restrictions from one dashboard.
Orgs, teams, rolesMulti-tenant by design: organizations, teams, and four role levels. Access enforced at the database layer with row-level security, not just in the app.
Semantic search, Cmd+KA command palette that searches meaning, not keywords, across your team's conversations. Keyboard-first, everywhere in the app.
Native desktop appmacOS and Windows via Tauri: system tray, native notifications, deep links for invites, auto-updates. Same UI as the web, no feature gap.
How deployment works
Start at step 01 →01Single container installOne Docker image, one Helm chart. Deploys to your cluster in under an hour, including SSO and storage configuration.
02Identity-native setupSAML and OIDC SSO with SCIM provisioning out of the box. Your users sign in with existing accounts; admin roles map to your IdP groups.
03Observability and handoffPrometheus metrics, OpenTelemetry traces, and structured audit logs plug into your monitoring on day one. We stay on call for the first 90 days.
Proof · Engineering firm, 200 employees
Replaced unmanaged ChatGPT use with a self-hosted workspace. Shadow AI dropped to zero; adoption did the opposite.
94%monthly active adoption
200employees onboarded
<4moto org-wide rollout
Self-hosted · SSO/SCIM · 200+ models · Team RAG · Audit logs
Request a demoTwo-week pilot in your VPC. If it does not fit, you keep the runbook.
Read the full case→
More products
All products →What a demo looks like
Book it →01Architecture call, 30 minWith our solution lead, focused on your stack and constraints, not a feature tour.
02Pilot proposal in 48 hoursScope, timeline, integration matrix, and success metrics, in writing.
03Reference call, optionalTalk to a team currently running Lumen in production before you commit.
Request a demo
“Shadow ChatGPT is not a policy problem, it is a tooling problem. Give people a workspace that is faster than the browser tab and remembers what the team knows, and the data stops leaving on its own.”
Sebastian Mondragon, Founder of Particula Tech
Request a demo30 minutes · no sales deck