UNPKG

eve

Version:

Filesystem-first framework for durable backend AI agents that run anywhere.

github.com/vercel/eve

vercel/eve

22 lines (19 loc) • 1.27 kB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
## Collect intent If the user's intent is not already clear, ask the questions one at a time, use the coding harness's prompt tools when available, and do not guess. 1. What should the agent do? This becomes its always-on purpose. 2. Where should it be reachable? Every agent ships the built-in HTTP channel. On top of that: - **Web Chat** (a Next.js app): add it at init with `--channel-web-nextjs`. - **Slack** and other platforms: add after deploy with `eve add channel/slack`. Credentials run through **Vercel Connect**, which provisions the bot token and verifies inbound webhooks, so there is no `SLACK_BOT_TOKEN` or signing secret to manage. 3. Which external systems does it need programmatic read/write access to, such as Slack, Salesforce, Linear, GitHub, or your own API? Each becomes a connection under `agent/connections/`. When a system needs every end-user to sign in, wire its auth through **Vercel Connect** (`connect()` from `@vercel/connect/eve`), which handles consent, encrypted token storage, and refresh. 4. Does the user want a specific AI Gateway model or reasoning effort? If so, pass its `provider/model-id` to `eve init --model` and its reasoning effort to `--reasoning`; otherwise, use eve's defaults.