V2 #1061
V2 #1061
Conversation
fix: optimize Twitter login with cookie validation and retry mechanism - Check for existing cookies or logged-in status before attempting login. - Add a retry mechanism with a limit of 5 attempts to avoid infinite loops. - Log meaningful errors and outcomes for easier debugging. - Reduce unnecessary API calls to improve efficiency and prevent rate limits.
fix: code fence alignment
feat: add venice.ai image generation
…eck-#855 feat: improve X/Twitter login with cookie validation and retry mechanism
feat: create README_VI.md
Update docs (CONTRIBUTING.md)
feat: add README_TH.md in docs directory
Codecov ReportAttention: Patch coverage is
|
|
For message bus - consider long term that we may want a way to coordinate different sovereign agents to interact with each other, so any design should consider this possibility. The message bus as scoped is internal to the agent for v2, but maybe there's just a little bit of design that we can do upfront to more easily support cross-agent coordination in the future. Here's some detail considerations to consider (from Sonnet FYI): Message namespacing that could later accommodate agent IDs |
|
SOPS might be a good tool for secrets management: https://github.com/getsops/sops btw, we are cooking a more advanced TEE plugin for Eliza that also uses SOPS for secure initialization: https://github.com/comrade-coop/aapp-toolkit |
This is a draft PR for the V2 branch.
This is extremely work in progress.
Here are the objectives of V2:
Unified message bus and simplified clients
We'll be refactoring all clients to handle input and output only, with an event bus to hook in response handlers.
Clients will have dramatically less code, and agents will be able to send messages across clients. This will also enable autonomous agent to unify all input and outputs into a single stream of thought and action.
Unified agent wallet
The current wallet system reflects the way we use humans use wallets today, but the UX around network switching and wallet management is a pain. Instead, the agent will have a unified wallet, with access to all registered chains at once. No wallet switching, and tokens will all be handled by agents in a unified way.
CLI Tool
We want a create-react-app like experience for creating new projects, and we want CLI handlers for adding and removing plugins, opening chat interface in browser, maybe even deploying like Cloudflare Wrangler to some of our deployment provider friends. Should also handle secrets management and character / agent / conversation management.
Model Provider Registry
The current model provider system requires a pull request to core for every model provider. v2 will have a registry and override pattern so new code doesn't need to be added to core, and any provider can quickly override with a few lines of code to add integration without a pull request.
Extensible and generic core framework
The v1 is very opinionated about prompts, state and some features. v2 will be very extensible so you can create different kinds of agents and override state management and context composition. For advanced developers, this will make the framework far more powerful and flexible for different user cases.
Community plugins
We'll be moving community plugins out of core and adding requirements for plugin inclusion in core around test coverage, documentation and code maintenance and ownership.
Better secret management
Currently, secret management in .env and character files is risky, so considering better and more secure options for this
100% Test Coverage
v2 reflects what we want Eliza to be, and we feel that after these changes we should focus entirely on stability, security and cutting down on issues. We'll be aiming for 100% test coverage of core, adapters and primary components, and end-to-end tests for clients.