elizaOS Documentation

repository·develop·Indexed 29 days ago

https://github.com/elizaos/eliza

Documentation for elizaOS, featuring the ElizaOS Context Benchmark for evaluating LLM reasoning, the @elizaos/agent package with HTTP backend and x402 micropayment support, and a comprehensive Deployment Toolkit for Docker-based application rollout.

Tokens
1.9M
Snippets
3.5K
Records
9.4K
Agent score
97%

What's inside eliza

  1. Overview of voice-classifier-cpp

    develop

    The voice-classifier-cpp package is a standalone C library and a set of GGUF conversion scripts. It ports native voice-side classifiers to the elizaOS/llama.cpp fork's ggml dispatcher. This is intended to replace the current onnxruntime-node path used by voice services in plugins/plugin-local-inference/src/services/voice/.

    Supported classifiers include:

    • Voice emotion classifier: Provides 7-class basic-emotion soft probabilities.
    • End-of-turn (EOT) detector: Returns audio-side $P(\text{end_of_turn}) \in [0, 1]$. Note: Currently returns -ENOSYS until an upstream audio-turn model is pinned.
    • Speaker embedding encoder: Provides 256-dim WeSpeaker embeddings with a cosine-distance helper.
    • Diarizer: Uses pyannote-3 segmentation powerset labels.

    Important: Before modifying this directory, read AGENTS.md for the full port plan, including upstream pins, GGUF schemas, and integration steps.

  2. Overview of the tunnel-proxy service

    develop

    The tunnel-proxy is a Railway-hosted service that provides public URLs for Eliza Cloud tunnels. It acts as a bridge between public TLS-terminated hostnames and internal Headscale MagicDNS hosts.

    Mapping Logic: Public hostnames like eliza-<org>-<random>.tunnel.elizacloud.ai are mapped to internal hosts like https://eliza-<org>-<random>.tunnel.eliza.local.

    Security & Routing:

    • When TUNNEL_HOSTNAME_SIGNING_SECRET is configured, the proxy only routes hostnames matching the pattern: eliza-<orgpart>-<randomhex>-<expiry>-<signature>.
    • Arbitrary wildcard labels or root traffic return a 404 error.
    • The /health and /ready endpoints remain public to allow Railway and DNS smoke checks to function.
  3. Overview of doctr.cpp (ggml port of doCTR)

    develop

    doctr.cpp is a C++ port of Mindee doCTR built on ggml. It provides a high-performance implementation of two stages for document processing:

    1. Detection: Uses a db_mobilenet_v3_large backbone with a DBNet head to produce a probability map.
    2. Recognition: Uses a crnn_mobilenet_v3_small backbone with a BiLSTM and CTC head to produce per-crop logits.

    Note on Workflow: The C++ library only handles the forward pass. The TypeScript caller orchestrates the full pipeline: detection $\rightarrow$ crop $\rightarrow$ recognition $\rightarrow$ decode. Post-processing (DBNet contour to bbox and CTC greedy decode) is handled in TypeScript for runtime portability.

  4. Overview of Meeting Corpus Importers

    develop

    The meeting-corpus-importers package provides license-gated importer contracts for the ElizaOS Context Benchmark. It is a code-only package designed to handle the discovery and parsing of meeting corpora without downloading or bundling actual data.

    Key capabilities include:

    • Corpus Registry: A P0 registry containing license and citation metadata.
    • Manifest Generation: Deterministic generation of local-cache manifests.
    • Fixture Parsing: A parser that emits structured data including transcript, speaker-turn, source-stream, RTTM, and annotation-coverage.
    • Status Reporting: Honest reporting of missing or unavailable licensed corpora.

    This package supports the same manifest shapes used by real imports such as AMI, CHiME/DiPCo, LibriCSS, VoxConverse, DIHARD, MUSAN, WHAMR, and LibriMix, but uses small synthetic fixture annotations for testing purposes.

  5. Overview of @elizaos/plugin-feed

    develop

    @elizaos/plugin-feed connects an Eliza agent to the Feed prediction market game. It provides a GUI operator view and an HTTP proxy layer that allows the agent to interact with Feed markets, posts, chats, and team states.

    Key features include:

    • Embedded Dashboard: Integrates the Feed operator dashboard as a GUI view.
    • Automatic Authentication: Automatically handles Feed authentication by preferring the existing Steward/Eliza Cloud session JWT (STEWARD_AGENT_TOKEN). If unavailable, it falls back to FEED_AGENT_ID and FEED_AGENT_SECRET with automatic token refresh.
    • API Proxying: Proxies requests between the elizaOS API surface and the Feed backend for markets, social, messaging, and admin controls.
  6. Overview of the BinaryAlert malware detection system

    develop

    BinaryAlert is an open-source serverless AWS pipeline designed for real-time and retroactive malware detection. It monitors S3 bucket uploads via S3 event notifications and triggers AWS Lambda analysis.

    Key capabilities include:

    • Real-time Detection: Automatically triggers analysis upon file upload to S3.
    • YARA Rule Matching: Uses configurable YARA rules for string, regular expression, and conditional matching to detect various malware types.
    • Specialized Handling: Supports unpacking UPX-compressed files, deep analysis of archived content, and integration with CarbonBlack.
    • Alerting: Provides real-time notifications based on matching results.
  7. Overview of @elizaos/plugin-blocker

    develop

    @elizaos/plugin-blocker provides focus and distraction control for Eliza agents. It enables website blocking via a SelfControl-style hosts engine and supports macOS/mobile app blocking.

    Key components include:

    • Providers: websiteBlocker and appBlocker (both gate to contexts: ["focus", "automation"]).
    • Services: WebsiteBlockerService and AppBlockerService.
    • Persistence: Uses a Drizzle pgSchema('app_blocker') for storing rules, active sessions, and allow lists.
    • UI: A focus overlay view accessible at /focus via the FocusView component.

    Note on Actions: This plugin does not register the BLOCK action. The BLOCK umbrella action is managed by @elizaos/plugin-personal-assistant to handle owner gating, scheduling, and chat dispatch.

  8. Overview of Tails operating system

    develop

    Tails is a portable, privacy-focused operating system based on Debian GNU/Linux. It is designed to protect user privacy and avoid censorship by routing all internet traffic through the Tor network.

    Key features include:

    • No Trace: When the computer is shut down, Tails leaves no trace on the host machine.
    • Portable: Runs from a USB stick instead of the installed OS (Windows, macOS, or Linux).
    • Secure Defaults: Includes pre-configured applications for sensitive document work and secure communication.
    • Open Source: Free to download and verifiable by independent security researchers.
  9. Overview of HyperliquidBench

    develop

    HyperliquidBench is a reproducible benchmark suite designed to measure the operational competence of Hyperliquid perpetual trading agents. Unlike simple PnL tracking, it verifies an agent's ability to correctly route orders, cancels, transfers, and leverage changes.

    Key features include:

    • Verification-driven scoring: Uses a Base + Bonus - Penalty formula to reward breadth, discipline, and action composition while penalizing spam.
    • Dual tracks:
      • Coverage track: Judges the overall operational surface area.
      • HiaN (Haystack-in-a-Needle) track: Verifies long-context precision by requiring agents to execute exact instructions from noisy prompts.
    • Auditability: Every action is logged with submission payloads, HTTP acknowledgements, and websocket confirmations.
  10. Overview of Python-SortedContainers

    develop
    Python-SortedContainers is a high-performance, pure Python library providing sorted container data structures. It offers SortedList, SortedDict, and SortedSet, all maintaining $O(\log n)$ time complexity for insertion, deletion, and lookup operations. Because it is written in pure Python, it does not require a compiler or pre-built extensions, making it easy to install and use in environments where C extensions are difficult to manage.
  11. Overview of the AutoRCCar Project

    develop
    AutoRCCar is a Python-based autonomous remote-controlled car system designed for real-time video stream processing, neural network model training, and intelligent driving decision-making. It integrates video processing (via Raspberry Pi camera), object detection (using Haar cascades), sensor fusion (ultrasonic sensors), and hardware control (via Arduino) to enable end-to-end autonomous driving.
  12. Overview of @elizaos/plugin-finances

    develop
    The @elizaos/plugin-finances plugin provides an owner-facing finance dashboard for elizaOS. It enables tracking of balance summaries, transactions, and recurring charges. It integrates with payment sources (via Plaid/PayPal bridges) and includes features for CSV imports, spending summaries, and subscription detection.