Quora is QuoraLogic's self-hosted suite: single sign-on identity, real-time voice AI, event-driven automation, and native clients — running on hardware you control, with no third-party model vendor in the call path.
Authentik as the fleet OIDC provider with an LDAP outpost, and QuoraIAM on top of it: a purpose-built user registry holding roles, grants, contacts, companies, and agent profiles. Every other service calls QuoraIAM directly rather than going through the identity provider.
A phone receptionist that answers live calls: voice-activity detection with barge-in, cut-off answer recovery, and a distinct AI persona per dialed extension.
Calls, SMS, email, chat, and calendar collapse into one event stream, then into rules and multi-step flows that can wait days for a follow-up.
Three local model tiers — fast, reasoning, generation — plus GPU speech-to-text and text-to-speech shared by every service on the mesh.
Native clients with real SSO, SIP registration straight to the session border controller, video meetings, and chat — no browser shell.
Each service below is in production on the fleet today, with its own documented architecture, migration history, and honestly-recorded open gaps.
Asterisk ARI handles inbound calls over WebSocket and RTP while voice-activity detection manages barge-in and resumes an answer that got cut off. Each dialed extension maps to its own persona — separate prompt, separate voice — resolved live from the identity registry.
Six connectors — call events, assistant transcripts, SMS and missed-call webhooks, IMAP mail, XMPP chat, and CalDAV/CardDAV — feed a rule engine and a flow executor that can suspend for hours or days waiting on a follow-up event, then pick up exactly where it left off.
Built from scratch on top of Authentik: it syncs users, holds a roles and grants model, and stores contacts, companies, and agent profiles. Every fleet service reads identity from its REST API instead of querying the identity provider directly.
Two Flutter codebases sharing an architecture rather than a UI: hand-rolled OIDC with PKCE, SIP registration straight to the session border controller, LiveKit video for meetings, and local SQLite for chat and notes. Mobile keeps its hub-and-spoke navigation; desktop is a plain window app.
Asterisk call events, Telnyx SMS and missed-call webhooks, IMAP mail, XMPP chat, and CalDAV changes land on one normalised event bus.
The rule engine matches the event, then local models on the inference host handle transcription, intent, and reply generation.
Flows execute across services: answer the line with the right persona, send the follow-up, book the slot, open the task.
Every event feeds a queryable behavioural profile — written to your own storage and vectorised so any fleet service can ask what is going on.
Traefik terminates TLS for roughly thirty services across four domains. Backend traffic stays on a private Netmaker mesh, with the session border controller deliberately kept off it for isolation.
Transcription, reasoning, and speech synthesis all run on fleet-owned GPUs. Segmentation is enforced below the application: Mongo, Postgres, and Redis each sit on isolated container networks, and multi-tenant data is bounded by Postgres row-level security rather than application checks alone.
No external model vendor in the call or message path
Row-level security keyed on company, not just app logic
Isolated networks per datastore; one publish path only
Outages diagnosed, fixed, and written down the same day
Personas, routing, flows, and ingress all live in version-controlled config next to the services they drive. If you can read a diff, you can audit the whole fleet.
1# Every fleet service reads identity from QuoraIAM,2# never from the identity provider directly.3 4iam = QuoraIAM(5 base_url="https://iam.quoralogic.io",6 audience="quora-voice",7)8 9profile = iam.agent_profile(extension="204")10# -> persona, voice_id, grants, company_idConnectors keep collecting, flows keep resuming, and the behavioural profile keeps rebuilding — including the flow that has been patiently waiting since Tuesday.
You run the fleet
We run it, you own the data
For teams with their own stack
Tell us what you run today — phone system, mail, chat, calendars — and we will map it onto the Quora suite.