Self-Hosted
Your servers. Your rules. No data leaves your network. Deploy a full AI agent platform on your own infrastructure.
The Problem
Every prompt, every document, every conversation — sent to servers you don't control, in jurisdictions you didn't choose.
Your agents depend on someone else's API. Pricing changes? Service outage? You're at their mercy.
GDPR, CLOUD Act, industry regulations — cloud AI makes compliance a moving target. Self-hosting makes it simple.
The Solution
Pinchy deploys on your infrastructure — bare metal, VM, or cloud instances you control. The data stays where you put it.
On-prem, private cloud, or your own VPS. You decide where Pinchy runs.
Conversations, documents, agent memory — everything stays on your network. Period.
Use any LLM — OpenAI, Anthropic, Google, or local models via Ollama. Go fully air-gapped: no internet connection needed, no data leaves the server. For finance, healthcare, and public sector, this is a requirement.
Update when you want. Test before you deploy. No forced changes, no surprises.
How It Works
Pinchy ships as versioned Docker images on GitHub Container Registry. No build step, no custom Dockerfile — pull the tag you want and run.
$ # Download the v0.8.0 compose file
$ curl -fsSL https://raw.githubusercontent.com/heypinchy/pinchy/v0.8.0/docker-compose.yml -o docker-compose.yml
$ docker compose up -d
✓ pinchy started (http://localhost:7777)
✓ openclaw started
✓ db started
🦞 Open http://localhost:7777 — setup wizard creates your admin
Connect your LLM provider, invite your team, pick agent templates. No license key needed for open-source features. Enterprise-gated features (Groups, RBAC, Per-User Breakdown) unlock with a trial key — get one in seconds or book a call.
Hardened for Operators
Recent releases focused on the operator experience — the signals, defaults, and boundaries that make self-hosting Pinchy in a regulated environment less of a guessing game.
Versioned Docker images on ghcr.io/heypinchy/pinchy. No source build, no compile step. Pin a tag, pull, run. Update when you're ready.
Run Pinchy without TLS and every page shows a banner until it's fixed. Hard to ship to production by accident — no silent insecure deployments.
Local Ollama or Ollama Cloud, configured per agent. Sensitive agents stay on-prem; others can use frontier models. More →
Token spend, provider cost, cache hits — split by agent, user, system vs plugin. You see where the money goes before the invoice arrives. More →
Every tool call, every knowledge-base hit, every approval — logged with the user and agent identity. Same view across web UI, Telegram, and plugin actions. More →
Pair Pinchy with local Ollama and nothing crosses the Atlantic — no US provider in the loop, no CLOUD Act exposure. Compliance by architecture. More →
Full Control
All agent conversations, documents, and memory stored on your database. Back up, migrate, or delete — your call.
Connect any LLM provider or run local models. No vendor lock-in. Switch models without changing agents.
Scoped per-agent permissions, user groups, and per-plugin boundaries. You define who can reach which agent, and what that agent is allowed to do.
Pull new versions when you're ready. Test in staging first. Roll back if needed. No forced updates.
FAQ
Self-hosting keeps all data on your infrastructure — no third-party access, no transatlantic transfers, no vendor lock-in. You control the models, the data retention, and the security boundaries. For regulated industries or privacy-conscious companies, it's the only viable option.
A server with Docker, 2GB+ RAM, and network access. A $10/month VPS works for small teams. For fully offline operation with local models, you'll want a machine with a GPU. Pinchy deploys with a single docker compose up command.
Yes. Pinchy supports local models via Ollama. In this configuration, agents run entirely on your own hardware and zero data leaves your network — the entire AI stack runs air-gapped on your hardware. No internet connection required after initial setup.
ChatGPT Enterprise processes your data on OpenAI's servers and locks you into one model provider. Pinchy runs on your servers, supports any model (OpenAI, Anthropic, local via Ollama), and has no per-seat pricing. You trade some setup convenience for complete data sovereignty.
Pinchy ships as pre-built Docker images on GitHub Container Registry (ghcr.io/heypinchy/pinchy). A docker-compose deployment is the default — no license key is required to run it. The free Community tier (AGPL-3.0) self-hosts with no key. To unlock enterprise features (Groups, agent access control, basic RBAC), activate a JWT enterprise license key — a trial key (all Pro features, no credit card) or a paid Pinchy Pro key — issued per organization and entered via Settings → License or the PINCHY_ENTERPRISE_KEY env var. Validated locally; Pinchy never phones home.
Self-host Pinchy yourself in minutes, or book a call to talk it through. Your choice.
Or email us: info@heypinchy.com