Packages
Browse packages indexed by the registry.
Local whisper.cpp speech-to-text backend (via whisper-rs) for atomr-agents.
Tool + Skill adapters that expose atomr-agents speech-to-text inside the agent framework.
Higher-level voice-session abstraction (Live vs TurnBased) plus VAD and mic→stream pump for atomr-agents speech-to-text.
MockInference, deterministic strategies, in-memory stores, replay harness.
Tool trait, ToolSet, registry, permission spec, and tool-call parsers.
Output-side audio for atomr-agents text-to-speech: PCM->WAV/Opus encoders, cpal speaker output, and stream-to-speaker pumps.
Core types and traits for atomr-agents text-to-speech: TextToSpeech / SynthesisStream / RealtimeSession traits, Capabilities, SynthesisRequest, VoiceRef, AudioOutput, MockTextToSpeech.
ElevenLabs TTS backend (REST + WebSocket streaming + voice cloning + SFX) for atomr-agents.
Gemini Live API backend for atomr-agents (bidirectional voice agents).
Kokoro-82M ONNX TTS backend for atomr-agents (local).
MOSS-TTS local backend for atomr-agents (TTS / SFX / dialogue / voicegen / realtime).
OpenAI TTS backend (tts-1, tts-1-hd, gpt-4o-mini-tts) for atomr-agents text-to-speech.
OpenAI Realtime API backend for atomr-agents (bidirectional voice agents).
Piper ONNX TTS backend for atomr-agents (local).
Coqui XTTS v2 backend for atomr-agents (local, voice cloning + 17 languages).
Tool + Skill adapters that expose atomr-agents text-to-speech inside the agent framework.
Bidirectional voice conversation: STT ↔ agent ↔ TTS, with unified-realtime auto-detection.
Provider-agnostic web-search trait, request/hit types, and a deterministic mock provider for tests.
Brave Search WebSearch provider for atomr-agents.
SerpAPI (Google) WebSearch provider for atomr-agents.
Tavily WebSearch provider for atomr-agents.
Generic `Tool` wrapper that exposes any `WebSearch` provider to atomr-agents agents and harnesses.
Workflow actor: DAG of Steps over event-sourced state.
Membership, gossip, reachability, vector clocks, and split-brain resolvers for atomr clusters.
Adaptive load balancing using cluster-wide metrics for atomr.
Sharded entities for atomr — allocation strategies, rebalance, three-phase handoff, remember-entities.
Distributed pub/sub, cluster singleton, and cluster client patterns for atomr.
Layered HOCON-style configuration for atomr — `include`, `${path}` substitution, `${?ENV}`, dotted keys, arrays.
Lease-based coordination primitives for atomr.
Actors, supervision, dispatch, mailboxes, scheduler, FSM, event stream, and coordinated shutdown — the core of the atomr actor runtime.
Live web UI over a running atomr system — REST + WebSocket + embedded React SPA, Prometheus / OTLP exporters.
Pluggable service-discovery interface for atomr.
Convergent replicated data types (CRDTs) over a atomr cluster — counters, sets, maps, registers, replicator.
redb-backed DurableStore for atomr distributed-data.
FIX session-layer for atomr — logon/heartbeat/sequence/resend/gap-fill session FSM over atomr-streams Tcp + Framing, with a pluggable persistent sequence store.
Builder API for wiring a atomr system together — config, DI, setup hooks.
Multi-runtime GPU + remote inference as a supervised actor system on atomr. Single-crate rollup that re-exports per-runtime backends behind feature flags.
atomr-infer serve CLI binary. Boots the actor system, applies project-file [[deployment]] entries, and mounts the gateway.
Foundation crate for atomr-infer — typed errors, deployment value object, ModelRunner trait, batch primitives.
atomr-streams integration for atomr-infer; re-exports the atomr-accel-patterns blueprints (DynamicBatchingServer, InferenceCascade, ModelReplicaPool, FairShareScheduler, ModelHotSwapServer, SpeculativeDecoder, MoeRouter).
Python-resident GPU runtime wiring for atomr-infer. PythonGpuBridge + python-pinned dispatcher used by the vLLM and other GIL-bound runners.
Remote-provider infrastructure for atomr-infer — distributed rate limiter (CRDT), circuit breaker, retry/backoff, SSE parser, session lifecycle.
Runtime-agnostic actor topology for atomr-infer — gateway, request actor, dp-coordinator, engine-core, two-tier worker, placement, deployment manager, metrics.
Anthropic Messages API provider for atomr-infer — implements ModelRunner against api.anthropic.com, with cost table, error classification, and SSE streaming.
Pure-Rust transformer inference for atomr-infer via the candle crate family. Feature-gated; default-features-off compiles to a typed-error stub.
Direct CUDA kernel dispatch for atomr-infer via cudarc + atomr-accel-cuda. Feature-gated; default-features-off compiles to a typed-error stub.
Google Gemini provider for atomr-infer — implements ModelRunner against Vertex AI and AI Studio endpoints, with cost table, error classification, and SSE streaming.
LiteLLM proxy provider for atomr-infer — implements ModelRunner against the LiteLLM unified API gateway, fronting OpenAI-compatible endpoints for any provider LiteLLM supports.
Mistral.rs runner for atomr-infer. Wraps mistralrs::TextModelBuilder + Model behind the ModelRunner trait with token-streaming.