Skip to content

XD319/auto-talon

Repository files navigation

AutoTalon v0.1.0

English | 简体中文

CI Node.js pnpm License: MIT

Personal assistant workspace for day-to-day execution.

AutoTalon is for personal knowledge workers who want a long-lived assistant without giving up inspectability, local control, or cost awareness. talon tui is the default home for daily work: today, inbox, threads, memory review, and governed task execution all live in one workspace. The CLI remains the automation, diagnostics, and maintenance surface, while Feishu/Lark and webhook gateways provide formal external chat entry points into the same runtime.

Primary Entry Points

  • talon tui Daily workspace for conversation, inbox triage, thread follow-up, and memory-aware execution.
  • talon gateway serve-feishu External IM entry for Feishu/Lark when you want the assistant to live inside a chat workflow.
  • talon run / talon continue Scriptable execution and follow-up from the terminal for automation, batch work, and precise inspection.

What It Does

  • Opens a personal assistant workspace in talon tui with today/inbox/thread oriented workflows.
  • Supports formal chat ingress through Feishu/Lark and local webhook adapters.
  • Records task state, trace events, tool calls, approvals, and audit logs in a local SQLite workspace.
  • Gates risky tool use with policy and explicit approval flows.
  • Surfaces memory review in both TUI and CLI, including used-memory feedback and inbox-driven suggestions.
  • Provides layered memory (profile/project/working + experience_ref/skill_ref) for repeatable work.
  • Keeps runtime observation available through talon ops and CLI inspection commands.
  • Supports replay, smoke tests, eval reports, and maintainer release checks in source checkouts.

Demo

$ talon init --yes
Initialized .auto-talon workspace files.

$ talon tui
# Open the daily workspace.
# Start or continue a thread, process inbox items, and review memory suggestions.

$ talon task list
$ talon trace <task_id> --summary
$ talon audit <task_id> --summary
# Drop to CLI when you want precise inspection or automation.

Quick Start

Requirements:

  • Node.js >=22.13.0
  • Corepack enabled for source installs

Installed package:

npm install -g auto-talon
talon init --yes
talon provider setup openai --api-key "$OPENAI_API_KEY"
talon tui

Optional chat-platform entry:

talon gateway serve-feishu --cwd .

Source checkout:

corepack pnpm install
corepack pnpm build
corepack pnpm dev init --yes
corepack pnpm dev provider setup openai --api-key "$OPENAI_API_KEY"
corepack pnpm dev tui

Typical Flows

Daily work in TUI:

talon tui
talon ops

Bring the assistant into chat:

talon gateway serve-feishu --cwd .
talon gateway list-adapters

Automate or inspect from CLI:

talon run "review the changed files"
talon continue --last
talon task list
talon trace <task_id> --summary
talon audit <task_id> --summary

Local API or SDK integration:

talon gateway serve-webhook --port 7070

Validate providers:

talon provider list
talon provider promote
talon provider status
talon provider test

When To Use It

  • You want a TUI-centered personal assistant workspace with auditable execution history.
  • You want today/inbox/thread actions to stay close to your terminal workflow without reducing the product to one-off prompt execution.
  • You want an assistant that can move between TUI, CLI, and chat-platform entry points while sharing the same governed runtime, memory, approvals, and audit trail.
  • You need policy and approval behavior before allowing file or shell actions.
  • You want durable memory, skill recall, replay, and eval tooling around ongoing knowledge work instead of one-off prompts.

Positioning

AutoTalon is a local-first personal assistant product for individual operators and knowledge workers, backed by an inspectable runtime rather than a hosted black box. The user-facing promise is a low-cost long-term assistant with a primary TUI workspace, supported by CLI automation and diagnostics plus formal external chat entry through adapters such as Feishu/Lark. The core package stays intentionally small, and integrations are loaded only when their gateway command is used. Runtime observation remains available via talon ops, while talon dashboard is preserved as a compatibility alias.

Documentation

User docs:

  • docs/user/install.md
  • docs/user/quickstart.md
  • docs/user/commands.md
  • docs/user/replay-and-eval.md
  • docs/user/approvals.md
  • docs/user/skills.md
  • docs/user/gateway.md
  • docs/user/mcp.md
  • docs/user/config-reference.md

Developer docs:

  • docs/dev/architecture.md
  • docs/dev/module-boundaries.md
  • docs/dev/plugin-development.md
  • docs/dev/testing.md

Troubleshooting:

  • docs/troubleshooting/provider.md
  • docs/troubleshooting/sandbox.md
  • docs/troubleshooting/gateway.md
  • docs/troubleshooting/memory.md

Release Validation

corepack pnpm check
corepack pnpm dev release check

talon eval run, talon smoke run, and talon release check are maintainer diagnostic commands for source checkouts. Installed-package users should use talon doctor and talon provider test for workspace health checks.

About

CLI-first agent runtime with governance and traceability

Topics

Resources

License

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages