I built a collection of open-source Eve agents for AI based on the Vercel framework

AI agents become much more useful when they are designed for a specific job instead of trying to do everything. That is why I built EveAgents: a growing collection of open-source AI agents built for the Eve framework. Each agent has a defined role, a practical workflow, domain-specific instructions, safety guardrails, and optional integrations for the tools teams already use. In this article, I will show you how to deploy an agent on Railway and introduce ten agents from the collection that cover engineering, product, analytics, customer success, marketing, finance, productivity, and research.

What I wanted to build

I did not want these agents to be a collection of vague system prompts. Each Eve agent is a self-contained project with the files needed to understand, run, and customize it:

• instructions.md defines the role, workflow, tool policy, and guardrails.
• agent.ts configures the model through Eve.
• skills/ contains the agent’s focused domain playbook.
• examples/ provides a representative prompt for testing the workflow.
• .env.example documents the required environment variables without exposing secrets.
• README.md and SETUP.md explain how to use the agent.

The agents are MIT-licensed, so you can inspect the source, adapt the workflow, add your own integrations, and deploy the result in your own environment. You can browse the collection on EveAgents or explore the open-source repository on GitHub.

How to deploy an Eve agent on Railway

Before looking at the ten agents, here is the fastest way to put one online. I will use the SEO Growth Analyst as the example. It is a useful starting point because the standalone version can work with information you supply directly, while optional variants can connect it to services such as Similarweb, Local Falcon, Webflow, Notion, Slack, and Teams.

1. Open the agent and choose the standalone version

Start on the SEO Growth Analyst page and review the agent’s role, source files, integrations, and setup information. For a first deployment, I recommend the standalone agent. It gives you the base workflow without requiring integration-specific credentials. You can deploy an integration variant later.

2. Create an EveAgents API key

Create a key from the EveAgents API Keys page and copy it somewhere secure. Railway uses this key during the build to download the selected agent from the EveAgents registry. The key is not the same as your AI model provider key. Keep both values private and never commit them to a repository.

3. Launch the Railway template

Open the EveAgents Railway template and create a new Railway project. The template retrieves the selected agent, validates its files, builds it with Eve, exposes it as a long-running service, and configures the health check at:

/eve/v1/health

It also stores workflow state on a persistent volume mounted at:

/app/.eve/.workflow-data

4. Configure the environment variables

For a standalone SEO Growth Analyst deployment, use the following configuration:

EVE_AGENT_SLUG=seo-growth-analyst
EVE_INTEGRATION_SLUG=
EVEAGENTS_REGISTRY_URL=https://www.eveagents.dev/api/registry/v1
EVEAGENTS_API_KEY=your_eveagents_api_key
EVE_PROVIDER_API_KEY=your_model_provider_api_key
EVE_MODEL=openai/gpt-5.4-mini
ROUTE_AUTH_BASIC_USER=eve
ROUTE_AUTH_BASIC_PASSWORD=use_a_long_generated_password

Here is what each value does:

The template supports OpenAI, Anthropic, and Google models. Make sure the model and provider key match. For example:

• openai/gpt-5.4-mini requires an OpenAI API key.
• anthropic/claude-sonnet-4-6 requires an Anthropic API key.
• google/gemini-3.6-flash requires a Google AI API key.

5. Deploy and verify the service

Start the Railway deployment and wait for the build to finish. Railway should use the included /eve/v1/health endpoint to confirm that the service is healthy. Once it is live, keep the basic-auth credentials enabled. The Railway URL is public, so these credentials prevent someone else from freely using your deployed agent and consuming your model credits.

If you add an integration, return to the relevant agent variant on EveAgents and add every environment variable listed in its individual setup guide. Standalone agents and direct-credential channels can run entirely on Railway. Slack and MCP/OpenAPI connection variants currently need a separately configured Vercel Connect authorization path.

10 open-source Eve agents in the collection

The deployment flow is the same for every base agent: change EVE_AGENT_SLUG, keep the registry and authentication variables, and add integration credentials only when the selected variant requires them. Here are ten agents I think demonstrate the range of what this collection can do.

1. SEO Growth Analyst

The SEO Growth Analyst finds evidence-backed search opportunities and turns them into prioritized content and site recommendations. I built it to go beyond generic keyword lists. Its job is to connect search intent, business relevance, content gaps, and available evidence so that a team can decide what to create or improve next. Optional variants can work with Similarweb, Local Falcon, Webflow, Notion, Slack, Teams, or Eve.

Railway slug: seo-growth-analyst

2. Incident Response Commander

The Incident Response Commander helps investigate incidents, coordinate responders, and turn incomplete evidence into a clear recovery plan. It is designed for high-pressure situations where logs, alerts, issue updates, and team messages are spread across multiple systems. Variants are available for tools such as Datadog, Sentry, Honeycomb, Linear, Notion, Slack, Teams, and Discord.

Railway slug: incident-response-commander

3. API Reliability Investigator

The API Reliability Investigator diagnoses failing APIs by connecting requests, logs, traces, monitors, and application errors. Instead of jumping to a conclusion after seeing one error, the agent is structured to assemble an evidence trail and separate confirmed findings from hypotheses. Its integration options include Postman, Datadog, Honeycomb, Sentry, GitHub, Slack, and Teams.

Railway slug: api-reliability-investigator

4. Product Feedback Synthesizer

The Product Feedback Synthesizer turns qualitative feedback and product signals into evidence-backed themes and opportunities. This is useful when feedback is arriving through interviews, support conversations, internal notes, analytics, and issue trackers. The agent helps group recurring themes without flattening important differences between customers. Variants can connect with Notion, Linear, PostHog, Mixpanel, Slack, Teams, and Eve.

Railway slug: product-feedback-synthesizer

5. Feature Adoption Analyst

The Feature Adoption Analyst explains adoption, drop-off, cohorts, and unexpected product behavior without overstating causality. I wanted an analytics agent that could help teams reason carefully about product signals. It can organize the evidence, identify useful comparisons, and make the limits of the available data visible. Available variants include PostHog, Mixpanel, ClickHouse, Notion, Slack, Teams, and Eve.

Railway slug: feature-adoption-analyst

6. Customer Support Triage Agent

The Customer Support Triage Agent classifies incoming requests, looks for grounded answers, drafts responses, and prepares clean escalations. Its goal is not simply to reply faster. It is designed to distinguish what can be answered safely from what needs human review or escalation. Variants are available for Slack, Teams, Telegram, Twilio, Notion, Airtable, Linear, and Eve.

Railway slug: customer-support-triage-agent

7. Customer Onboarding Concierge

The Customer Onboarding Concierge guides customers through onboarding with contextual next steps, progress tracking, and timely follow-up. This agent is useful when onboarding involves multiple milestones rather than one welcome message. It can help keep the plan understandable, surface blockers, and adapt the next action to the customer’s current state. Integration variants include Notion, Airtable, Todoist, Mem0, Slack, Teams, Telegram, and Eve.

Railway slug: customer-onboarding-concierge

8. Revenue Operations Analyst

The Revenue Operations Analyst reconciles revenue, payment, expense, cash, and operational records into a decision-ready report. Finance work benefits from clear provenance and cautious conclusions, so the workflow is designed to organize records, surface discrepancies, and make unresolved questions explicit. Variants can connect with Stripe, Brex, Embat, Airtable, Slack, Teams, and Eve.

Railway slug: revenue-operations-analyst

9. Meeting Action Planner

The Meeting Action Planner turns rough meeting notes into a decision log and an owner-based action plan. This is one of the simplest agents to try because you can paste in a transcript or a page of notes and ask it to extract decisions, owners, follow-ups, open questions, and next steps. It is a focused workflow that solves a familiar problem without pretending that every sentence in a meeting is equally important.

Railway slug: meeting-action-planner

10. Nonprofit Grant Researcher

The Nonprofit Grant Researcher finds aligned funders, assesses eligibility, and maintains an evidence-backed grant opportunity pipeline. It is designed to help research teams distinguish a promising opportunity from one that only looks relevant at first glance. The agent keeps eligibility, deadlines, thematic fit, evidence, and open verification work visible. Variants are available for Candid, Notion, Airtable, Slack, Teams, and Eve.

Railway slug: nonprofit-grant-researcher

You can also run an agent locally

Railway is convenient when you want a persistent hosted service, but every standalone agent can also run locally. For example, you can install the SEO Growth Analyst with:

npx @bergside/eveagents install seo-growth-analyst
cd seo-growth-analyst
npx eve@latest

To see every available slug, run:

npx @bergside/eveagents list

This makes it easy to inspect the instructions and skill files before deciding what to customize or deploy.

Why I made the collection open source

Agent workflows should be inspectable. If an agent is going to help triage an incident, analyze revenue records, recommend content, or draft a customer response, teams should be able to see the instructions and guardrails behind its behavior. Open source also makes the collection more useful as a starting point: you can change the model, adapt the workflow, add a connector, or reuse the structure for a completely new role.

The result is not one general-purpose assistant with a long list of responsibilities. It is a collection of focused agents that can be understood, tested, and improved one workflow at a time.

You can browse all of the open-source agents on EveAgents, start with the SEO Growth Analyst, or deploy an agent using the Railway template.