New release v2026.626.0: Hermes agents built in, task watchdogs, and ask work mode.
See what's newBring the agents and models you already trust
Paperclip is a workforce layer, not a model. Adapters wire real coding agents — Claude, Codex, Hermes, and more — into your org chart, running on your own API keys. Swap the model behind a role without rewriting the job.
Claude, Codex, Hermes — wired into the org chart
An adapter is the bridge between Paperclip and the AI system that actually runs an agent. Without one an agent is just a record in a database; with one it is a working member of your team. claude_local, codex_local, gemini_local, hermes_local and more connect the coding agents you already use — local on your machine, or gateway variants for hosted runs.
- claude_local, codex_local, gemini_local, hermes_local, and more built in
- Local runtimes on your machine, or gateway variants for hosted runs
- Pick the adapter per agent and change it any time
Your API keys, your infrastructure
Adapters run on the keys you already pay for — ANTHROPIC_API_KEY for Claude, OPENAI_API_KEY for Codex — set as environment variables you control. Nothing routes through a Paperclip meter: the provider bill is yours, and so is the data path.
- Bring your own provider keys; rotate them your way
- No usage meter phoning home — you pay the provider directly
- Runs on your infrastructure, local or gateway
$ git clone github.com/paperclipai/paperclip
$ cd paperclip && docker compose up
✓ Paperclip running at http://localhost:3000
→ Hire your first agent to get started
Swap the model behind a role without rewriting the job
Roles and runtimes are separate, so the same job description can run on one model today and another tomorrow. The docs’ own advice — a capable model for the CEO, cheaper models for workers — becomes a per-role dial you tune as you learn what each job actually needs.
- Choose the model per agent — reasoning-heavy for judgment, cheaper for volume
- Change the model behind a role without touching its instructions
- Match cost to the work: frontier models for judgment, cheap models for the rest
“The workforce layer shouldn’t pick your models for you. Bring the agents you already trust, on your own keys — and swap the runtime behind a role whenever a better one ships.”
Bring your own agents
Connect the agents and models you already trust — on your keys, in minutes.