One key, one base URL, three interfaces. The HawkTalk models — the Hawk Alpha family — served on AWS Inferentia, reachable over REST, WebSockets, and HawkTalkLive. Everything here is the same auth, the same models, the same trace format.
Every request is authenticated with a bearer key from your dashboard. REST + WS take it as a header or a session.start field; keep it server-side.
Base URL https://api.hawktalk.ai Header Authorization: Bearer $HAWKTALK_API_KEY
Tiny-to-capable, all function-calling native, all runnable on-device and served in the cloud. Pick the smallest one that clears your task; escalate up the family (or up the Blade Frontier) only when you need to.
| model id | role | good for |
|---|---|---|
| hawk-sapling | smallest · fastest | Classification, routing, tight function-calling. Sub-second on the edge. |
| hawk-feedseed | grounded / retrieval | RAG, extraction, answers pinned to your data. |
| hawk-sakura | balanced · conversational | Everyday voice + chat, the default for assistants. |
| hawk-oak | largest · most capable | Complex reasoning, multi-tool agents, the hard turns. |
Voices are separate from models — pass voice: "af_heart" (and others) on any streaming tier.
| interface | endpoint | use it for |
|---|---|---|
| REST | POST /v1/generate | One-shot text + function-calling. Stateless, everywhere. |
| WebSockets | wss /v1/stream | Token-by-token + streamed voice, barge-in. "Realtime." |
| HawkTalkLive | wss /v1/live | Reactive-state voice — text, tools, thinking & async on independent channels. |
| plan | model access | throughput | live tier |
|---|---|---|---|
| Pro $120/yr | Sapling → Sakura | Standard · batched under load | — |
| Dev $600/yr | Full family, incl. hawk-oak | Higher concurrency + priority · generous credits | Capped rates |
| Freedom $3000/yr | Full family | Flat per-1M-token · dedicated dl2q · no caps | Included |
Full trace logs on every plan. See pricing for the breakdown.
Standard HTTP status + a typed JSON body. Retriable errors set retry_after.
| status | code | meaning |
|---|---|---|
| 400 | invalid_request | Malformed body or unknown field. |
| 401 | unauthorized | Missing or bad API key. |
| 404 | unknown_model | That model id isn't in the family. |
| 429 | rate_limited | Over your plan's throughput — back off retry_afters. |
| 503 | capacity | Tier saturated; retry or step down a model. |
Every response — REST, WS, or Live — carries a trace_id. Pull the full, human-readable trace of exactly what ran (tokens, tool calls, channel activity) from the dashboard or the API. Built on mkmdmcp principles: nothing the engine does is a black box.
curl https://api.hawktalk.ai/v1/traces/trc_c14a90 \ -H "Authorization: Bearer $HAWKTALK_API_KEY"