Alan AI工作流引擎

Prerequisites

• Rust 1.85+ (2024 edition)
• just (task runner, optional but recommended)

Building

```bash git clone <repo-url> cd alan cargo build --release

Installation

The supported macOS distribution is app-first. The signed Alan.app bundle contains the alan CLI and alan-tui executable under Contents/Resources/bin.

```bash

Normal user install

brew install --cask alan

Local developer install from this checkout

ALAN_DEVELOPER_ID_APPLICATION="Developer ID Application: Example (TEAMID)" just install

Quick Start

CLI Usage

```bash

API Usage

Route ownership lives in crates/alan/src/daemon/api_contract.rs; the examples below show the stable public paths, while production clients should use the contract helpers or generated TUI helpers.

```bash

Create response sample fields

Configuration

The recommended setup path is launching alan chat or alan-tui and using the first-run wizard. The wizard starts with user-facing service presets such as ChatGPT/Codex login, OpenAI API Platform, OpenRouter, Kimi Coding, DeepSeek, Google Gemini via Vertex AI, and Anthropic API. Raw API-family selection is kept behind Advanced / custom setup.

Connection/provider metadata lives in ~/.alan/connections.toml; agent runtime knobs live in ~/.alan/agents/default/agent.toml. An agent config can pin a profile with connection_profile, otherwise alan resolves a workspace pin or the global default_profile.

```toml

Optional explicit pin

Optional skill exposure overrides

[[skill_overrides]] skill = "plan" allow_implicit_invocation = false

[[skill_overrides]] skill = "release-checklist" enabled = false

Optional explicit compaction budgeting override

Thinking / reasoning (optional)

model_reasoning_effort = "medium" ```

Host-facing daemon/client settings live in ~/.alan/host.toml. You can also set ALAN_CONFIG_PATH to use a custom agent config file location.

Policy Configuration (Optional)

Create {workspace}/.alan/agents/default/policy.yaml to override builtin policy profile rules. When present, the file replaces the builtin profile rule set for that session; alan does not implicitly merge policy files with builtin rules.

rules:
  - id: deny-prod-delete
    tool: bash
    match_command: "kubectl delete"
    action: deny
    reason: protect production cluster

  - id: review-prod-deploy
    tool: bash
    match_command: "deploy --prod"
    action: escalate
    reason: explicit production boundary

default_action: allow

See docs/governance_current_contract.md for the current implementation contract. HITE target-design material now lives in OpenSpec under openspec/specs/governance-tooling-contract/spec.md.

---

"websocket_url": "/api/v1/sessions/.../ws",

"events_url": "/api/v1/sessions/.../events",

"submit_url": "/api/v1/sessions/.../submit",

Poll events from rollout gap-aware API

curl "http://localhost:8090/api/v1/sessions/{id}/events/read?after_event_id=e-123&limit=50"

Skill catalog and override APIs

curl http://localhost:8090/api/v1/skills/catalog curl "http://localhost:8090/api/v1/skills/changed?after=<cursor>" curl -X POST http://localhost:8090/api/v1/skills/overrides \ -H "Content-Type: application/json" \ -d '{"skill_id": "memory", "allowImplicitInvocation": false}'

Inspect resolved skills, packages, package exports, and availability

alan skills list alan skills packages alan skills init ./my-skill --name my-skill alan skills validate ./my-skill --strict alan skills eval ./my-skill

One-shot question

alan ask "What does this project do?" alan ask "Explain main.rs" --workspace ./my-project alan ask "Summarize" --output json # NDJSON for automation alan ask "Summarize" --output quiet # text only at end alan ask "Think step by step" --thinking --timeout 60