dylanneve1/talon
Live in productionRun persistent AI agents across Telegram, Discord, Teams, terminal and desktop—with multiple backends and full MCP tool access.
🦅 Multi-platform agentic AI harness — runs on Telegram, Discord, Teams & Terminal with a pluggable backend (Claude, Kilo, OpenCode, Codex, OpenAI Agents), full MCP tool access, and persistent background agents (Goals, Heartbeat, Dream).
- TypeScript84.2%
- Dart10.6%
- JavaScript1.9%
- Rust1.3%
- Go1.3%
- Kotlin0.3%
- C0.2%
- Zig0.2%
1 Review
Talon is an ambitious and impressively complete agent platform. It brings several AI backends together behind one system and supports Telegram, Discord, Microsoft Teams, the terminal, and a Flutter companion app. The architecture is carefully separated into core, backend, frontend, storage, and plugin layers. This is a strong choice for a project with so many moving parts because it limits direct dependencies between features. The shared protocol fixtures are another excellent decision. Testing the same messages across the daemon, companion app, and headless node should help catch compatibility problems before they reach users.
The repository also shows unusually strong attention to quality and security. It includes more than 2,300 tests, cross-platform checks, live backend tests, fuzz testing, CodeQL, dependency review, secret scanning, Docker validation, and release package checks. The security policy gives users a private reporting path and clearly explains bridge authentication, encryption, file permissions, and the project’s main threat areas. Recent CI, CodeQL, release, and application builds are passing. One Dependabot update run failed, but related dependency pull requests passed their main CI and security checks, so this appears to be maintenance activity rather than a sign that the project is unstable.
The biggest improvement would be making the first-time experience simpler. The README is detailed, but it introduces many advanced systems before a new user has seen one complete example. A short “choose your path” section could guide users toward terminal-only use, a messaging bot, or the companion app. Each path should list its exact requirements, setup commands, and a quick way to confirm that it works. A small diagram showing how the frontend, Talon daemon, AI backend, and MCP tools connect would also make the system easier to understand.
There is one documentation mismatch worth correcting. The README badge says Node.js 22 or newer, while the quick-start section and package.json require Node.js 24 or newer. Keeping those values consistent will prevent avoidable setup failures. It would also help to explain the support level of each backend and platform, including which combinations receive full live testing.
Because Talon can access files, messages, websites, and background tasks, safe defaults are especially important. The existing security work is strong, but a dedicated threat model would make it even better. It could explain prompt injection risks, tool approval rules, plugin trust, credential handling, and the limits of background agents. Overall, Talon is a serious, well-engineered project with excellent testing and a clear long-term vision. Its next major win would be turning that technical depth into a calmer and more guided onboarding experience.
