A conversational front-end that routes user intent to the right Arcanada service: ask Verdicus to translate, Transcribator to convert audio, Munera to track a task. Planning phase, target port 3800. Distinct from Argana AI (commercial product) — ARCA Assistant is the operator-facing ecosystem control plane, sharing the @arcanada/core libraries.
Capabilities
- Natural-language intent routing across all Arcanada services
- Cross-service orchestration (one prompt → multiple service calls)
- Auth Arcana session — every action is identity-attached
- Per-user memory via LTM (recall preferences across sessions)
- Web + Telegram clients planned
- Shared @arcanada/core with Argana AI (DRY connector layer)
- Audit trail of every routed intent for post-hoc review
Current autonomy level
L3
Weakest link
Not yet in production. The L4 self-healing capabilities of downstream services do not help if ARCA Assistant itself misroutes intents — intent-classification evaluation set is the gating artifact.
Roadmap to L3
- L1 → L2 — MVP launch with manually curated intent map; deploy reporting to Ops Bot.
- L2 → L3 — observability stack, intent-classification eval set (≥500 labeled samples), fallback to fail-fast on unknown intent rather than guessing.
- Verification gate — eval set ≥ 95% accuracy and zero silent misroutes for 30 days post-launch.