Victoi's companion.
Probing /actuator/health…
Probing POST /mcp…
19 verified · 0 provisional · 19 total · anthropic (1) gemini (5) imagen (2) veo (2) lyria (2) bge (1) e5 (1) nllb (1) wav2lip (1) codeformer (1) real-esrgan (1) bopbl (1) /v1/drivers → in docs →
| id | family | capabilities | status |
|---|---|---|---|
anthropic/claude-sonnet-4-6 |
anthropic | streamtext.generatetool.usevision.read | verified |
google/gemini-3.5-flash |
gemini | streamtext.generatetool.usevision.read | verified |
google/gemini-3.1-pro |
gemini | streamtext.generatetool.usevision.read | verified |
google/gemini-embedding-2 |
gemini | embed | verified |
google/imagen-4-fast |
imagen | image.generate | verified |
google/imagen-4-ultra |
imagen | image.generate | verified |
google/nano-banana-pro |
gemini | image.editimage.generatetext.generatevision.read | verified |
google/veo-3-fast |
veo | video.generate | verified |
google/veo-3-pro |
veo | video.generate | verified |
google/lyria-3-clip |
lyria | music.generate | verified |
google/lyria-3-pro |
lyria | music.generate | verified |
google/gemini-3.1-flash-live |
gemini | audio.inaudio.outlive.bidistreamtext.generatetool.use | verified |
local/bge-reranker-base |
bge | embed | verified |
local/multilingual-e5-base |
e5 | embed | verified |
local/nllb-200-distilled |
nllb | text.generate | verified |
replicate/wav2lip |
wav2lip | lip.sync | verified |
replicate/codeformer |
codeformer | image.face_restoreimage.restore | verified |
replicate/real-esrgan |
real-esrgan | image.upscale | verified |
replicate/bringing-old-photos-back |
bopbl | image.restore | verified |
| MCP tools registered | 22 |
|---|---|
| Organs in blueprint | 12 (implemented: 11) |
| Status page | GET / |
| Beautiful docs | GET /docs · Scalar via CDN |
| OpenAPI | GET /openapi.json |
| MCP | POST /mcp · JSON-RPC 2.0 (initialize / tools/list / tools/call) |
| Health | GET /actuator/health · /info · /metrics |
Returns Viki's identity: name, version, persona, values and brand truths. Static — owned by the system, not by any vendor model.
Returns the capability taxonomy (text.generate, vision.read, audio.in / audio.out, live.bidi, embed, image.generate, video.generate, music.generate, search.grounded, etc). Each capability lists which registered drivers fulfil it — empty in v0 because the driver catalogue is empty.
Lists Viki's organ blueprint (Identity, NervousSystem, Ears, Eyes, Mind, Hands, Mouth, Voice, Imagination, Tongue, Hippocampus, Skin). Each entry carries its responsibility, the capabilities it requires from a driver, and a v0 status flag.
Returns the registered Driver catalogue and the JSON schema each Driver entry must satisfy. Each driver carries a ``probe_path`` pointing at the live capture that backs its metadata. Pass ``capability`` (e.g. 'text.generate') or ``vendor`` to filter.
Returns the architecture-v0 doc as markdown. Without arguments, returns the full document plus a per-section breakdown. Pass ``section`` to receive just one section's body (prefix-matched against the section title, case-insensitive).
Returns the full Driver record, the live-probe capture that backs it (raw request + response), and the generated RAG markdown card. The single best way to learn exactly how a model behaves on Viki.
Returns a Victoi product knowledge-base document from vikii/knowledge/products/ as markdown. Without arguments, lists the available products (slug, doc_id, title, category). Pass ``product`` — accepts either the doc_id (e.g. 'payments-rails', 'jangi-savings-circles') or the filename slug — to fetch one document. Add ``section`` to narrow to a single section (prefix-matched against the section title, case-insensitive — e.g. 'faq', 'use cases', 'how it works', 'key features').
DEPRECATED — use `vikii.identity` instead. Returns Viki's single identity wrapped in a one-element list for back-compat with the pre-2026-05-23 multi-persona shape.
Returns the intent taxonomy the IntentClassifier picks from — chitchat, question_simple, question_complex, lookup, search, transactional, generative_text, generative_media, support, voice_live, educational, urgent. Each entry includes example utterances, the keyword signals the rule-based classifier matches on, the default task-kind sequence, and whether the intent is side-effecting.
Multimodal media → text. Pass `media_base64` + `media_kind` ('audio' / 'image' / 'video') and optionally `task` ('transcribe' / 'describe' / 'extract' / 'understand'). Returns the model's text output. Same payload as POST /v1/inputs/understand over REST.
Returns the full platform-tool catalogue (Hands organ) — what Viki can DO on Victoi. Each entry includes category, risk level, JSON schemas, backing service and compliance notes. Filter with `category` (payments / market / core / knowledge / memory / search / system) or `risk` (read / write.user / confirm / escalate / external).
Returns the full spec for one platform tool — schema, backing service, risk, compliance notes.
Run one platform tool against the supplied (or fresh) session. Returns the result AND the activity row that landed on the audit trail. Side-effecting tools (risk=confirm) refuse without a confirmation token.
Returns a single typed snapshot of the user's identity, Victoi account state (fiat balances + stock positions), conversation state, and memory pointers. Fans out to core-api, payments-service, market-data-service in parallel with JWT forwarding; in-process TTL cache (30s) per (user, session). Pass the returned snapshot to vikii.skill.thinking.reason to prime the Mind organ.
DEPRECATED (Sprint 4 / G5) — prefer ``vikii.subagent.invoke`` with ``subagent='ears'`` for the transcribe op. This bridge stays available for callers that haven't migrated; ``op='config'`` is kept here. Two ops in one tool: `op='transcribe'` runs an audio blob through Gemini multimodal STT (returns text + latency); `op='config'` returns the tuned VAD / endpointing / semantic-turn / streaming-TTS settings the LiveKit voice pipeline applies (audit A.5).
DEPRECATED (Sprint 4 / G5) — prefer ``vikii.subagent.invoke`` with ``subagent='voice_session'`` for ``mint_session``. The other ops (interrupt_ack / filler / streaming_tts_enabled) remain here until they land as Mouth subagent ops in Sprint 6. Voice-side composition. Ops: `mint_session` — issue a LiveKit access token + queue dispatch; `interrupt_ack` — pick a verbal barge-in ACK phrase for the language (audit A.6); `filler` — pick a filler clip metadata (by kind or by tool name); `streaming_tts_enabled` — read the streaming-TTS feature flag (audit A.4).
Register an already-uploaded attachment as a Vikii document and enqueue parsing + embedding. Returns the doc_id callers can poll for ingest_status via vikii.document.list.
List the caller's uploaded documents with their ingest status. Soft-deleted docs are excluded unless include_deleted=true.
Top-K semantic search across the caller's uploaded documents. Optionally scoped to specific doc_ids. Returns snippets with doc:{doc_id}#p{page} citations.
Soft-delete one of the caller's documents. Chunks remain in storage for the auto-purge window (90 days for customer scope) so undelete is possible until then.
Long-form structured summary of one of the caller's documents. Returns tldr + key_findings + risks + cited_chunks. Runs via DEEP_THINK so cost scales with tier (Claude Sonnet on pro, Opus on premium).
Generate a Sulafat-voiced monologue audio briefing for one of the caller's documents. Returns attachment_id (playable via GET /v1/attachments/{id}), duration, and the transcript. Per-tier daily quota applies.
Vikii v0 is the foundation. Drivers, providers, organs and the nervous system
arrive incrementally — see vikii.describe_architecture via the
MCP for the target shape, or open the API docs for the
current HTTP surface.