Help shape the next terminal-first CLI commands

[happy520ai]

The first command set has shipped

Unified AI System is terminal-first: developers can script, inspect, and operate the gateway without depending on a browser UI. The CLI, HTTP API, shared SDK, MCP server, and public container form the product path.

Try the isolated CLI without cloning the repository:

docker run --rm ghcr.io/happy520ai/unified-ai-system/ai-gateway-service:latest pnpm gateway demo

The v0.2.0 source workflow includes:

Command	Current job
pnpm gateway demo	Run one isolated fake-provider request and clean up.
pnpm gateway serve	Start the self-hosted gateway.
pnpm gateway status	Inspect health, provider mode, and chat readiness.
pnpm gateway chat	Send one request with a real-provider authorization boundary.
pnpm gateway doctor	Check Node, pnpm, workspace, and gateway connectivity.
pnpm gateway version	Print the CLI version.

demo, status, chat, doctor, and version support --json. CLI chat sends no request when a real provider may be active unless the operator explicitly adds --allow-real-provider for that command.

The latest public-tree cleanup retired the legacy /ui and /console routes, removed 168 generated historical artifacts, and added checks that keep the default runtime terminal-only.

Choose the next complete path

Candidate areas:

1. models and routes explain for provider selection and routing evidence
2. agent plan and agent run with approval checkpoints
3. knowledge add, knowledge search, and retrieval diagnostics
4. approvals list and approvals decide for human control
5. mcp setup and connection diagnostics for developer clients
6. A lightweight TUI built on the same command and JSON contracts

Which workflow would make you use the gateway for a real local project? Concrete commands, expected JSON fields, and failure cases are especially useful.

Design boundaries

• Reuse the gateway and shared SDK; do not create a second runtime.
• Keep the deterministic fake provider as the no-credential default.
• Keep real providers and cost-bearing requests explicit opt-in.
• Preserve approvals, permissions, evidence, and process cleanup.
• Treat production readiness, L5 autonomy, and AGI as evidence gates, not marketing language.

CLI reference | English README | Chinese README | Roadmap

[happy520ai]

v0.2.0 Terminal CLI Preview is live

The first working command set has now shipped:

pnpm gateway demo
pnpm gateway serve
pnpm gateway status
pnpm gateway chat "Hello from the terminal"
pnpm gateway doctor

demo, status, chat, doctor, and version support --json. The chat command also fails closed when a real provider may be active: it sends no chat POST unless the operator authorizes that one command with --allow-real-provider.

Release and exact clone instructions: https://github.com/happy520ai/unified-ai-system/releases/tag/v0.2.0

The next CLI surface should be chosen by actual operator value. Feedback is especially useful on which area should come first: agent plans, knowledge/RAG, model routing, or human approvals.