​
Install & setup

Yes. Self-hosted is free forever under FSL‑1.1‑ALv2. All features, unlimited agents, unlimited users. The license only restricts re-selling Qorven as a hosted service that competes with ours.
  • Linux or macOS, x86_64 or arm64
  • ~2 GB RAM, ~5 GB disk
  • PostgreSQL 14+ with pgvector (installer handles it)
  • An LLM provider key (or Ollama for local)
  • That’s it.
Yes — a Pi 5 with 8 GB handles single-user just fine. Use Ollama with a small model (Qwen 2.5 7B, Llama 3.2 3B) for zero-network operation.
Either. The installer script targets bare metal (systemd); docker compose up works identically. Teams deploying to Kubernetes should see docker + reverse proxy.
WSL2 (recommended) or Docker Desktop. Native .exe runs but needs manual PostgreSQL setup + a service wrapper (NSSM). The binary itself compiles fine.

​
Models & costs

Any OpenAI-compatible API: OpenAI, Anthropic, Bedrock, DeepSeek, Gemini, Groq, Mistral, Together, Fireworks, Ollama, LM Studio, Perplexity, DashScope, OpenRouter. Custom providers can register via the provider interface.
DeepSeek Chat for default + Gemini Flash for long-context tasks runs most users under $1/month. Fully local with Ollama is free (you pay in hardware).
Only the LLM calls you authorise. If you pick OpenAI as your provider, yes — you’ve asked Qorven to send prompts to OpenAI. That’s the only external traffic. No telemetry, no training, no phone-home.
Yes. Set type = "bedrock" with your AWS profile. Claude, Llama, Nova, Titan, and others work.
Models overview + qorven models search. 1,700+ models with live pricing from providers, refreshed weekly.

​
Privacy & data

Only LLM provider calls (to your configured provider) and web-fetches your Qor explicitly makes. Every outbound destination is in the audit log.
No. Zero. The binary doesn’t call home, doesn’t report usage, doesn’t check for updates without you running qorven update.
PostgreSQL + /var/lib/qorven. All on your infrastructure. The encryption key in config.toml protects provider keys + OAuth secrets + vault entries.
Yes. Run air-gapped with Ollama (no outbound LLM calls), use customer-managed Postgres, enable PII redaction, lock down tenant isolation. Enterprise compliance packet available on request.
qorven memory purge --user <id> + qorven sessions delete --user <id>. Soft-deletes propagate to backups after retention expires. Privacy →

​
Usage

ChatGPT is one cloud assistant. Qorven is many agents on your machine that delegate to each other. Your data stays yours. You pick the LLM.
Those are libraries — you write Python to configure them. Qorven is a self-hosted product with a UI, channels, memory, audit. You don’t write a workflow; you ask Prime and Prime delegates.
Yes. Rooms are multi-agent collaboration spaces. Prime can delegate to one specialist or drop 3 into a room for coordinated work.
Same thing. “Qor” is Qorven’s branded term, short for Qorven Operating Role. We prefer it because “agent” is overloaded.
No hard cap. Qors are rows in agents, not processes. One install with 100 Qors uses the same memory as one with 10 Qors — they’re only active when handling a message.

​
Where next

Debugging

Logs, doctor, common failure modes.

Getting help

Discord + GitHub Issues + commercial support.