Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

64 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Runyard Crew

Open-source AI Management System for your company.

Agents, skills, memory, and policies live as files in a git-like project workspace you own. Work runs in local session branches. Finished work lands through human-approved change requests. Built on Eve — Vercel’s filesystem-first durable agent framework.

Self-host free. Target 1 is a customer-owned Railway template with BYO infrastructure and model keys. Runyard Cloud comes later as a managed service. See the launch decision.

Status

Local command center is runnable on Eve + Next. Demo data persists under .crew-data/. CI runs typecheck + Eve evals (.github/workflows/ci.yml).

Shipped (local / stub-complete)

Area What works today
P2 git OS Projects as git repos, session branches, change requests, Review
P3 workforce Agents/skills, templates, marketplace install stub, enable/disable
P4–P5 ops Secrets, connectors + broker, approvals, cron/webhooks, idle reap
P6 sandbox Docker + Vercel through Eve; Railway checkpoint adapter; Runyard preview bridge; provider capability matrix and crew sandbox
P7 channels/CLI Slack/Discord/Teams inbound stubs, crew CLI (projects/sessions/billing/sandbox/llm/auth/tenant), Compose self-host
P8 billing Seats, wallets, Stripe checkout + HMAC webhook, credit holds, budgets/BYOK, crew billing
P9 enterprise OIDC PKCE auth + rate limits, SCIM Users, audit export, Postgres world, evals/CI, dedicated SKU, LLM route + token-proxy plan, org pin (CREW_ORG_ID)

Still stub / commercial follow-through

  • Live Stripe Dashboard + IdP token exchange (HMAC verify + OIDC authorize/callback stubs are ready)
  • Live Runyard SSE/spawn and enforced Sandbox egress (native bridge meters policy as a no-op)
  • Multi-org request routing (process pin + optional .crew-data/orgs/<orgId> shard — docs/multi-tenant.md)
  • Live LLM upstream token proxy + Braintrust eval dashboards (plan/stub proxy is ready — docs/llm-gateway.md)

Brand and licensing split: docs/brand-split.md. Cloud plans: docs/cloud-sku.md.

Stack

  • Runtime: Eve ^0.31 (durable sessions, tools, skills, channels, sandbox)
  • UI: Next.js 16 (proxy.ts request gating) + withEve / Web Chat command center at /projects
  • Workflows: Eve local world by default (.eve/.workflow-data); Postgres via DATABASE_URL / @workflow/world-postgres for self-host/Railway (see docs/world.md)
  • Sandbox: Docker locally, Railway Sandboxes on Railway, Vercel Sandbox on Vercel, and Runyard preview via @runyard/sandbox-provider
  • Evals: Eve evals/ harness (npm run eval) — smoke placeholders skip without model keys (see docs/evals.md)

Prerequisites

  • Node.js 24.x
  • Docker (for local sandbox / Compose Postgres)
  • A model credential (AI_GATEWAY_API_KEY or provider keys)

Deploy on Railway

Railway cannot use the local Docker sandbox. Configure a Railway project token for the native Railway Sandbox adapter plus a model provider before chatting with the agent. See docs/railway.md for the complete service, volume, security, and template profile.

Quick start

git clone https://github.com/runyard-dev/runyard-crew.git
cd runyard-crew
npm install
cp .env.example .env   # add AI_GATEWAY_API_KEY or provider keys

# Optional: Postgres for durable workflow world
docker compose up -d
# npm run world:bootstrap   # with DATABASE_URL / WORKFLOW_POSTGRES_URL set

# Seed demo project (also happens automatically on first API access)
npm run seed

# Web command center (Next + Eve)
npm run dev

Open http://localhost:3666/projects. Demo project: Acme Ops (acme-ops).

CLI (local stub)

npm run crew -- projects list
npm run crew -- projects create --name "Ops Desk" --template ops
npm run crew -- sessions new acme-ops --prompt "Draft weekly report"
npm run crew -- sessions list acme-ops
npm run crew -- sessions stop <sessionId>
npm run crew -- audit export --format jsonl -o /tmp/crew-audit.jsonl
npm run crew -- billing status
npm run crew -- billing topup --wallet aiCredits --amount 100
npm run crew -- sandbox status
CREW_RUNYARD_SDK_STUB=1 npm run crew -- sandbox probe
npm run crew -- llm status
npm run crew -- llm proxy --alias fast --message "ping"
npm run crew -- auth status
npm run crew -- tenant status
npm run crew -- world status --ping
npm run eval -- --list
npm run crew -- cr list acme-ops --status open
npm run crew -- skills install acme-ops standup-notes
npm run crew -- templates list

See docs/cli.md. Talks to the same .crew-data/ store as the web UI (does not run the agent loop).

Self-host (Docker Compose)

Postgres-only (for local npm run dev):

docker compose up -d

Full stack (Postgres + web image):

docker compose --profile full up -d --build
# or
npm run crew -- self-host up

See docs/self-host.md.

Command center routes

Route Purpose
/projects Project list + create
/projects/start Create wizard
/projects/[id] Home composer — start a session
/projects/[id]/sessions/[sessionId] Eve chat + Actions / Files / Audit / Approvals
/projects/[id]/files Repo file tree
/projects/[id]/agents Agents from crew.yaml
/projects/[id]/skills Skill docs
/projects/[id]/connectors Connectors
/projects/[id]/triggers Triggers (toggle)
/projects/[id]/review Change requests merge/close
/projects/[id]/settings Project settings
/approve/[id] Approve / deny (public deep link — see docs/auth.md)
/billing Seats, wallets, Stripe stub, budgets — see Billing UI + docs/cloud-sku.md + docs/stripe.md
/login Auth/SSO stub — docs/auth.md
/audit Org audit export — docs/audit.md
/api/crew/world Workflow world status — docs/world.md
/api/crew/scim/v2 SCIM Users stub — docs/scim.md
/api/crew/llm · /api/crew/llm/proxy LLM backends (AI Gateway / OpenRouter / Cloudflare) + token-proxy plan — docs/llm-gateway.md
/api/crew/tenant Org pin / isolation stub — docs/multi-tenant.md
/api/crew/billing/stripe/webhook Stripe HMAC webhook — docs/stripe.md
/api/crew/auth · /api/crew/auth/oidc/callback Auth + OIDC — docs/auth.md
/api/crew/projects/[id]/channels/* Slack / Discord / Teams inbound stubs — docs/channels.md
/api/crew/hooks/[triggerId] Webhook triggers (X-Crew-Webhook-Secret) — docs/triggers.md

Docs

Doc Topic
docs/sandbox-runyard.md Runyard SDK sandbox hook
docs/sandbox-providers.md Docker, Vercel, Runyard, E2B, Daytona, and Cloudflare provider contract
docs/functional-parity-roadmap.md End-to-end functional parity milestones
docs/cli.md crew CLI commands
docs/llm-gateway.md AI Gateway / OpenRouter / Cloudflare backends + token-proxy plan
docs/stripe.md Stripe checkout webhook HMAC
docs/multi-tenant.md Org pin / Cloud isolation notes
docs/brand-split.md Crew OSS vs Cloud / Sandbox
docs/self-host.md Docker Compose
docs/world.md Postgres Workflow world
docs/auth.md Login / SSO env
docs/scim.md IdP provisioning
docs/audit.md Compliance export
docs/channels.md Chat channel stubs
docs/triggers.md Cron + webhook triggers
docs/cloud-sku.md Cloud plans including Dedicated
docs/evals.md eve eval harness

Data lives in .crew-data/ (gitignored). Installed Eve docs: node_modules/eve/docs/.

License

Apache License 2.0

Related

Product Role
Crew (this repo) AI team / management system
Runyard Sandbox Hardware-isolated microVM compute
Runyard Cloud Managed Crew + Sandbox

About

Runyard Crew — open-source AI Management System. Company agents, skills, and memory as files in git; durable Eve agents; change-request review. Built on the Eve framework.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages