explaingit

second-inc/second

★ 26TypeScriptAudience · ops devopsComplexity · 4/5ActiveLicenseSetup · moderate

TLDR

Open-source internal-tools platform where humans and AI agents share the same MongoDB-backed apps, with bring-your-own runtimes like Claude Code or Codex.

Mindmap

mindmap
  root((second))
    Inputs
      App description
      Agent runtime
      MongoDB data
    Outputs
      Internal app
      Scheduled jobs
      Audit logs
    Use Cases
      Build research dashboard
      Lead enrichment pipeline
      Multi-agent workflows
      Self-host on Kubernetes
    Tech Stack
      TypeScript
      MongoDB
      Claude Code
      Codex
      Kubernetes

Things people build with this

USE CASE 1

Spin up an internal competitor-research dashboard that humans and agents both edit

USE CASE 2

Run a lead enrichment pipeline where one agent researches and another sends alerts

USE CASE 3

Self-host the platform on an air-gapped Kubernetes cluster with per-app permissions

USE CASE 4

Let a non-technical PM describe an app in English and have Second scaffold it

Tech stack

TypeScriptMongoDBNode.jsKubernetes

Getting it running

Difficulty · moderate Time to first run · 30min

Local try-it currently requires an Apple Silicon Mac; Intel Mac, Linux, and Windows builds are not yet shipped.

Apache 2.0 lets you use, modify, and redistribute the code commercially as long as you preserve attribution and the license notice.

In plain English

Second is an open-source platform for building internal software at companies, with a focus on letting AI agents work alongside human teammates inside the same apps. The project is at the npx try-it stage, currently shipping for Apple Silicon Macs, with Intel Mac, Linux, and Windows listed as coming soon. The pitch is that most workplace tools were not designed for AI agents, so they end up either ignoring them or bolting them on. Second tries to treat agents as first-class users from the start. The basic idea is that a non-technical team member can describe an internal app they want, for example a competitor research dashboard or a lead enrichment pipeline, and Second builds it. Once the app exists, AI agents read and write to the same real-time MongoDB database as the human team, get tools that are scoped specifically to the app, and can run scheduled or long-running jobs in the background. Users bring their own agent runtime. Claude Code and Codex are supported today, and OpenCode is marked as coming soon. The README says you can switch runtimes and models per message or per app without restarting the conversation. Several agents can also run in parallel on the same app, for example one handling research while another sends alerts and a third does enrichment. The product targets two audiences. Teams get a no-code surface for building apps and a shared workspace where humans and agents collaborate live. Platform engineers get one-time workspace setup, draft and review and publish lifecycles, fine-grained permissions per app and per agent and per integration, audit logs, and the option to deploy on their own Kubernetes cluster, on-prem, or air-gapped. The license is Apache 2.0. The README also points enterprise customers toward a sales contact for SSO, runtime setup, cost management, and SLA support.

Copy-paste prompts

Prompt 1
Try Second locally on an Apple Silicon Mac via npx and walk me through wiring Claude Code as the runtime
Prompt 2
Define an internal app spec in Second for a CRM enrichment workflow with two parallel agents
Prompt 3
Set up Second's draft, review, and publish lifecycle with per-agent permission scopes
Prompt 4
Compare deploying Second on Kubernetes vs running it on-prem in air-gapped mode
Prompt 5
Build a tool inside Second that lets an agent read from and write to a shared MongoDB collection
Open on GitHub → Explain another repo

Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.