1. Project Overview
eliza is an open-source TypeScript framework for building autonomous AI agents โ software entities that can chat, remember context, use tools, and act across platforms like Discord, Telegram, and blockchains without requiring developers to write integration code from scratch.
2. Background & Positioning
elizaOS was created to give developers an accessible, extensible foundation for building AI agents that operate consistently across many surfaces โ web, desktop, mobile, and messaging platforms โ while preserving user autonomy and, where desired, privacy through local, on-device inference. The core mission is to make agent development as approachable as writing a plugin, rather than as building a bespoke integration for every platform and model provider.
Unlike single-purpose chatbot SDKs or closed agent platforms, elizaOS is:
- Runtime-centric, not app-centric โ a shared
@elizaos/coreruntime handles memory, the message loop, and agent state, so the same agent definition can run as a Discord bot, a web app, or a CLI process. - Model-agnostic โ agents can call cloud model providers or run local models (2Bโ27B parameters) through the Eliza-1 model registry, giving teams a choice between convenience and full local control.
- Deeply pluggable โ nearly all agent capability (platform connectors, blockchain wallets, tools) is delivered through a large, community-maintained plugin ecosystem rather than being hard-coded into the framework.
3. Feature Categories
- ๐ Platform Connectors โ 90+ plugins, examples: Discord, Telegram, Slack, Twitter/X, Farcaster. Purpose: let a single agent definition operate natively across chat and social platforms.
- โ๏ธ Blockchain & Wallets โ examples: Ethereum plugin, Solana plugin, non-custodial wallet actions, transaction approval flows. Purpose: allow agents to hold and use crypto wallets under explicit, bounded user approval.
- ๐ง Model & Inference โ examples: Eliza-1 local model registry, cloud model provider adapters, Eliza Cloud model routing. Purpose: give developers a choice between local, private inference and hosted, cloud-routed inference.
- ๐งฉ Agent Runtime Primitives โ examples: Actions, Providers, Evaluators, Services, Characters. Purpose: the composable building blocks every plugin registers into, forming the agent's behavior and personality.
- ๐๏ธ Productivity & Utility Integrations โ examples: calendar access, reminders, document/knowledge processing, automation tools. Purpose: extend agents beyond conversation into practical daily-task assistance.
4. Key Highlights
- Character-driven personalities โ agent behavior and tone are defined declaratively through "character" files, separating personality configuration from runtime code.
- Actions, Providers, and Evaluators model โ a clear separation between what an agent can do (actions), what context it can pull in (providers), and how it assesses its own responses (evaluators), making plugin behavior predictable and composable.
- Local-first inference option โ the Eliza-1 model registry lets agents run entirely on-device, which matters for teams with privacy or offline requirements, alongside an optional cloud path.
- Non-custodial blockchain support โ wallet operations on EVM and Solana chains are scoped with explicit approval boundaries rather than unrestricted transaction signing.
- Multi-platform application suite โ beyond the framework itself, elizaOS ships web, desktop, mobile, and even bootable OS distributions, so the same agent can be deployed to end users, not just developers.
- Large first-party plugin ecosystem โ over 90 plugins cover messaging platforms, blockchains, and device bridges, reducing the amount of custom integration code most projects need to write.
5. Use Cases by Role
- General developers โ build a custom chat or automation agent by composing existing plugins (Discord, Telegram, calendar) instead of writing platform SDKs by hand.
- Data / research scientists โ use knowledge and document-processing workflows combined with local model inference to experiment with retrieval and memory-augmented agents without sending data to third-party APIs.
- Project managers โ deploy an agent for calendar management, reminders, and cross-platform coordination (e.g., a Discord/Slack assistant) with minimal custom code.
6. Getting Started
Find what you need โ browse the plugin registry and documentation:
https://docs.elizaos.ai/
Install / integrate โ clone the repository and start the development environment:
git clone --filter=blob:none https://github.com/elizaos/eliza.git
cd eliza
bun install
bun run dev
Contribute โ open an issue to discuss non-trivial changes, then submit a pull request against the develop branch:
git clone https://github.com/elizaos/eliza.git
See CONTRIBUTING.md in the repository for full requirements.
7. Project Structure
eliza/
โโโ packages/
โ โโโ core/ # @elizaos/core โ runtime, types, message loop, memory primitives
โ โโโ agent/ # @elizaos/agent โ standalone agent process and HTTP backend
โ โโโ app-core/ # @elizaos/app-core โ application hosting and orchestration
โ โโโ ui/ # @elizaos/ui โ shared React components
โ โโโ cli/ # command-line tooling for scaffolding and running agents
โโโ plugins/ # first-party plugins (platform connectors, blockchains, models)
โโโ scripts/ # repository maintenance utilities
โโโ patches/ # dependency patches
The packages/core runtime is the shared foundation every other package and plugin builds on; plugins/ is the primary place developers add or discover new agent capabilities.
8. Related Ecosystem
- Eliza Cloud โ an optional hosted platform providing authentication, model routing, and deployment for agents built with the open-source framework.
- Eliza-1 model registry โ a set of local models (2Bโ27B parameters) that plug into the runtime's model-handler interface for on-device inference.
- Model providers and messaging platforms โ the plugin system connects to third-party services such as Discord, Telegram, Slack, and blockchain networks (Ethereum, Solana), which remain independently operated services outside the elizaOS project.
9. License
- โ Use, modify, and distribute eliza, including for commercial purposes, under the MIT license.
- โ Fork the project and build proprietary agents or products on top of it.
- โ Do not remove the original copyright and license notice from copies or substantial portions of the software.
- โน๏ธ MIT provides no warranty; review the license text before relying on eliza in production, and check individual plugin licenses if they connect to third-party paid services.
10. FAQ
Q: Do I need to use Eliza Cloud to run an agent?
A: No. Eliza Cloud is optional; agents can run fully locally or against your own configured model providers.
Q: Can agents run without sending data to any cloud model provider?
A: Yes, via the Eliza-1 local model registry, which supports on-device inference for models in the 2Bโ27B parameter range.
Q: How do I add support for a new chat platform?
A: Build or install a plugin exporting a Plugin object that registers the relevant actions, providers, and services โ see the plugin documentation at docs.elizaos.ai.
Q: What blockchains does eliza support out of the box?
A: First-party plugins currently cover EVM-compatible chains and Solana, with non-custodial wallet operations subject to explicit approval boundaries.
Q: Where do I report a bug or propose a feature?
A: Open an issue in the GitHub repository before submitting non-trivial pull requests, per CONTRIBUTING.md.
11. Quick Links
- Repository: https://github.com/elizaOS/eliza
- Documentation: https://docs.elizaos.ai/
- Contributing guide: https://github.com/elizaOS/eliza/blob/develop/CONTRIBUTING.md
- Issues: https://github.com/elizaOS/eliza/issues
12. Summary
eliza gives developers a model-agnostic, plugin-driven runtime for building AI agents that work consistently across chat platforms, blockchains, and local or cloud inference backends. It is most valuable for teams that want to ship a multi-platform agent quickly without hand-rolling integrations for every messaging service or model provider, and for developers who want the option of fully local, private inference.