try.akko-ai.com — public read-only sandbox¶
try.akko-ai.com is the public, anonymous sandbox of the AKKO platform.
It lets a prospect explore two of AKKO's flagship features — natural
language to SQL through the AI layer, and document question answering
through the RAG layer — without any sign-up, login, or credential.
Read-only by design
The sandbox is deliberately minimalistic. It ships only what actually works in anonymous mode (3 cockpit pages : Home, AI/ADEN, RAG). Everything else (Notebook, Dashboards, NORA, Catalogs, Governance, AI Models, Logs, Alerts, Architecture) is hidden.
What you can do¶
| Feature | What happens |
|---|---|
| AI layer (ADEN) | Ask a question in natural language. The platform generates SQL against a curated demo catalog (banking, healthcare, retail), runs it through the Query layer, and renders the result as a chart + table. Nine verified queries are pre-seeded with cached responses for instant feedback. |
| RAG layer | Search two seeded collections (regulatory-docs, product-docs). The platform retrieves the top relevant chunks via vector search, generates an answer through the AI gateway, and returns the answer + citations. |
| Cockpit navigation | Browse the three enabled pages — Home, AI, RAG. The sandbox banner and diagonal watermark are visible on every page. |
What you cannot do¶
| Restriction | Why |
|---|---|
| No upload | Sandbox is shared and ephemeral. Uploading documents requires the full platform. |
| No write to the SQL editor | ADEN runs only the verified queries. Custom SQL is rejected with HTTP 403. |
| No catalog creation, deletion, or modification | Catalog mutations are an admin path and require RBAC. |
| No notebook / dashboard / governance | Those layers require a real identity and a real tenant. |
| Session expires after 30 minutes | Hard cookie TTL enforced at the Identity middleware layer. |
| All data resets every hour | A scheduled job wipes the sample data and re-seeds the demo. |
Quotas¶
| Limit | Value |
|---|---|
| Requests per second per IP | 10 (burst) |
| Requests per minute per IP | 100 (sustained) |
| ADEN runs per hour per IP | 50 |
| RAG queries per minute per IP | 5 |
| Query timeout | 30 s |
| Query scan budget | 500 MB |
| Upload body size | 1 MiB |
| Session TTL | 30 min |
| Egress to public internet | blocked |
When a quota is exceeded the platform returns HTTP 429 with a Retry-After header. The cockpit surfaces a sandbox toast pointing to the contact form.
Architecture¶
The sandbox is its own isolated Kubernetes namespace (akko-try) with
its own pods for every layer (Storage, Query, AI, RAG, Cockpit). It
does NOT share any state with the production AKKO instance. The only
cross-namespace traffic is an egress connection from the RAG layer to
the shared AI gateway (LiteLLM), which is explicitly allow-listed via
NetworkPolicy.
flowchart LR
User[Prospect on Internet]
Caddy[Caddy frontal]
Traefik[Traefik routing]
Cockpit[Cockpit + Cockpit Backend]
ADEN[AI layer ADEN]
RAG[RAG layer]
PG[(Storage layer postgres-try)]
LLM[AI gateway LiteLLM shared]
User -->|HTTPS| Caddy --> Traefik
Traefik -->|/api| Cockpit
Traefik -->|/| Cockpit
Cockpit -->|/api/aden| ADEN --> PG
Cockpit -->|/api/rag| RAG --> PG
RAG --> LLM
ADEN --> LLM
Every public-facing request goes through eight Traefik middlewares : strip-untrusted (anti-spoof), auth-inject (canonical identity), two rate-limit layers (burst + sustained), body-limit (1 MiB), session-ttl (30 min), watermark-hdr (sandbox header), cors-strict (single origin allow-list).
Try it¶
Sign up for the full platform¶
Want the full AKKO platform — your own catalogs, your own dashboards,
your own AI agents, your own governance ? Get in touch via the
contact form or write to
contact@akko-ai.com.
Spec references¶
- Sprint plan : sprint-86-A-bis-akko-try-chart
- Architecture decision : ADR-066 — autonomous chart for the public sandbox
- Helm chart :
helm/akko-try/(standalone, not a sub-chart)