| name | platform-knowledge |
| description | Qorven platform self-knowledge — architecture, capabilities, installable plugins, and self-evolution guide. |
| when_to_use | When asked about what the platform can do, what tools are available, how to extend capabilities, or when suggesting improvements. |
| context | inline |
Qorven Platform Self-Knowledge
Architecture
- Backend: Go (port 4200) — agent loop, tools, memory, supervisor
- Frontend: Next.js (port 3000) — Discord-style chat UI
- LLM Proxy: LiteLLM (port 4100) — routes to 13+ models
- Database: PostgreSQL + pgvector — memories, sessions, agents
- STT: Whisper small (port 8881) — voice transcription
- TTS: Edge TTS (free) + ElevenLabs/OpenAI (paid)
Available Tools
| Tool | Purpose | Cost |
|---|
web_search | Find information (Tavily, Perplexity, DuckDuckGo) | Free |
web_fetch | Read a specific URL | Free |
firecrawl | Deep crawl entire sites, JS rendering | Free if self-hosted |
exec | Run shell commands in sandbox | Free |
file_read/write/edit | Workspace file operations | Free |
email_send | Send email via SMTP | Free |
email_read | Check IMAP inbox | Free |
memory_search | Search agent memories | Free |
delegate_to_soul | Delegate task to another agent | Free |
create_soul | Create a new specialized agent | Free |
Installable Plugins
Plugins extend Qorven's capabilities. Install via Docker or API key:
Firecrawl (Web Crawling)
- API: Set
CRAWL4AI_API_TOKEN=fc-xxx (from firecrawl.dev)
- Self-hosted:
docker run -d -p 3002:3002 ghcr.io/firecrawl/firecrawl:latest
- Then set
FIRECRAWL_URL=http://localhost:3002/v1
Scrapling (CSS Scraping)
- Python-based CSS selector extraction
- Install:
pip install scrapling
- Use for: extracting specific elements from pages
SearxNG (Private Search)
- Self-hosted meta-search engine
- Install:
docker run -d -p 8888:8080 searxng/searxng
- Then set
SEARXNG_URL=http://localhost:8888
Self-Evolution Capabilities
Qorven agents can extend the platform:
- Install plugins — agents can suggest and guide plugin installation
- Create new agents — use
create_soul to spawn specialized workers
- Build skills — write SKILL.md files for reusable workflows
- Add connectors — 50+ integrations available, agents know how to configure them
- Research & learn — agents use web_search + firecrawl to stay current
Knowledge Base Architecture
- Company memories: Shared across all agents (decay-exempt)
- Agent memories: Per-agent, with certainty levels (explicit > deductive > inductive > abductive)
- Session memories: Per-conversation, auto-compacted
- Knowledge graph: Entity extraction from conversations
- Vector search: pgvector embeddings for semantic retrieval
When Users Ask About Capabilities
If a user asks "can you do X?" and the answer is "not yet":
- Check if a plugin exists for it
- Suggest installation steps
- Offer to create a skill for it
- If it requires code changes, explain what would need to be built
Outbound Actions
All outbound actions (email, social posts, webhooks) go through an approval gate:
- none: Auto-send (trusted agents only)
- supervisor: Prime reviews before sending
- user: Human approves in the UI
- both: Prime + human both approve
