Pinchy vs AnythingLLM
AnythingLLM is a polished, self-hosted document-chat and RAG app: a free desktop app, a free Docker multi-user mode, and a managed cloud, with a no-code agent builder on top. Pinchy is a governed multi-user agent platform: per-agent tool allow-lists, roles (and groups in Pro), and an HMAC-signed, hash-chained audit trail in the free Community edition. Both are open source, and both run on your own machine. They optimize for different jobs.
The Core Difference
Point it at your files, pick a model, and ask questions. The center of gravity is local document Q&A: a clean desktop app for one person, plus a Docker mode and a hosted option for small teams, with an agent builder layered on.
Pinchy runs on your own infrastructure and treats governance as the product. Each agent gets an explicit tool allow-list, every user has a role, and every tool call lands in a signed audit trail. Agents do not just answer; they act, under control.
AnythingLLM is MIT and Pinchy is AGPL-3.0, so neither locks you in at the code level. The difference is the bet: AnythingLLM optimizes for simple local document chat, Pinchy for governed team agents that take action and leave a record.
Side by Side
Figures as of 16 October 2026, taken from AnythingLLM's public pricing, product and documentation pages. Check AnythingLLM for current details.
| AnythingLLM | Pinchy | |
|---|---|---|
| Primary focus | Document chat / RAG | Governed multi-user agents |
| License | MIT | AGPL-3.0 |
| Free tier | Yes (free desktop app; Docker self-hosting is free too) | Yes, Community edition (AGPL), no user limit |
| Hosting | Desktop, Docker, managed cloud | Self-hosted (Docker Compose) |
| Local models / offline | Ollama and local LLMs | Ollama, fully offline |
| LLM providers | Many (30+ providers) | Anthropic, OpenAI, Google, Ollama |
| No-code agent builder | Yes (visual agent flows) | No visual builder |
| Single-user desktop app | Yes (Mac, Windows, Linux) | No (server app) |
| Channels | Web UI, chat widget (Docker), Telegram bot (single-user mode only) | Web UI + Telegram, also in multi-user instances |
| Business-system actions | SQL, web, files, MCP servers; Gmail and Outlook agents (single-user mode only) | Odoo read-write, email (Gmail, Microsoft 365, IMAP), web search |
| Per-agent tool allow-list | Skills switched on instance-wide (Settings > Agent Skills); per workspace only model and provider change | Empty allow-list per agent; admin grants each tool and directory |
| Roles | Admin / Manager / Default (Docker) | Admin / Member free; groups and agent visibility in Pro |
| Audit trail | Event logs (logins, messages, settings, uploads); no signing or chaining documented | Every row HMAC-SHA256-signed and hash-chained, built-in verifier, CSV export, free |
| MCP support | Yes | No |
| Granular RBAC / SSO | Roles (Docker); simple SSO passthrough self-hosted; SSO and RBAC listed under Enterprise | On roadmap, not shipped |
| Scheduled runs | Yes, Scheduled Jobs (single-user mode only) | No, agents act on request |
| Telemetry | Anonymous telemetry (opt-out) | Zero telemetry, no phone-home |
Being Honest
If your goal is to drop in a folder of PDFs and start asking questions, AnythingLLM nails it. Its RAG pipeline works out of the box with almost no setup, and it accepts far more file formats than Pinchy's index, which reads text-based PDFs. Pinchy has a knowledge base of its own since v0.9.0 (folders granted per agent, answers cited to the page), but for pure document chat, AnythingLLM is still the simpler path.
AnythingLLM ships a native desktop app for Mac, Windows, and Linux. For a single user who just wants a local AI on their own machine, that is a friendlier starting point than running a server stack.
It includes a visual, no-code agent flow builder and MCP support, so you can wire up web browsing, SQL, charts, and external tool servers without code. Pinchy has no visual builder and no MCP support.
In single-user mode, AnythingLLM's agent can search, read and send Gmail or Outlook mail (it asks for approval before actions that modify the mailbox), run Scheduled Jobs on a cron schedule, and answer through a Telegram bot. For one person who wants a personal assistant, that goes further than Pinchy, which has no scheduled runs.
AnythingLLM connects to a very wide range of LLM and embedding providers. Pinchy is model-agnostic too (Anthropic, OpenAI, Google, Ollama), but AnythingLLM's provider list is longer if exotic backends matter to you.
Where Pinchy Wins
Pinchy agents get no tools beyond their own workspace until an admin grants them (default-deny); templates propose a starting set, and only admins create agents. Admins grant each tool and directory explicitly, per agent: for example email_draft but not email_send, or Odoo read tools without write tools. In AnythingLLM, skills are switched on for the whole instance in Settings > Agent Skills; per workspace you change only the model and provider.
AnythingLLM keeps event logs of logins, messages, settings changes and uploads; no signing or chaining is documented. In Pinchy every tool call lands in an audit row that is HMAC-SHA256-signed and hash-chained, with a built-in verifier and CSV export. It is tamper-evident. For a CISO who needs to prove what an agent did, that record is the point.
AnythingLLM's agents act too, but its Gmail and Outlook skills, Scheduled Jobs and Telegram bot work only in single-user mode; multi-user mode hides or disconnects them. In Pinchy, Odoo (read-write), email over Gmail, Microsoft 365 or IMAP, web search and Telegram are per-agent grants inside a multi-user instance, and every call lands in the audit trail.
Pinchy is built for teams from the start: Admin and Member roles and user invites in the free edition, plus Groups and personal, shared or restricted agents in Pro. Governance applies across the whole team, not just to one person's workspace.
With a local model via Ollama, prompts and documents stay on your network; only the external integrations you grant (web search, Gmail or Microsoft 365, Telegram) reach outside. With a local model and no external integrations, it runs fully air-gapped. Pinchy has zero telemetry and validates its license offline, so it never phones home. API keys are AES-256-GCM encrypted at rest.
The allow-list per agent, the signed audit trail, the integrations and offline operation are in the free Community edition with no user limit. Pro adds groups, agent visibility and usage analytics (€1,188 per year for up to 10 users). The license covers neither your servers nor your model usage. AnythingLLM is free to self-host too; its hosted cloud starts at $50 per month.
Decision Guide
You mainly want to chat with your own documents, you want a polished desktop app for a single user or small setup, and the no-code agent builder plus broad provider support fit how you like to work. It also fits when you want a personal assistant on your own Gmail, Outlook or Telegram, with scheduled jobs.
You need governed agents for a team: per-agent permissions, roles (groups in Pro), a signed, hash-chained audit trail, and email, Telegram and Odoo that work for the whole team, not only in single-user mode. It means running a Docker Compose stack and choosing a model; there is no desktop app.
Is it one person's assistant, or a team's agents? If it is "chat with my files", or one user acting through their own Gmail, that points to AnythingLLM. If it is "governed agents that do things for a team, with a record of every tool call", that points to Pinchy.
FAQ
Yes. AnythingLLM by Mintplex Labs is open source under the MIT license and public on GitHub. You can run it as a free single-user desktop app on Mac, Windows, or Linux, or self-host the multi-user version with Docker for free. Mintplex also offers a managed cloud option. Pinchy is also open source (AGPL-3.0) and self-hosted first, with a single Docker Compose stack as the primary way to run it.
It depends on what you need. AnythingLLM is excellent if your goal is to chat with your own documents locally. Pinchy is a governed multi-user agent platform: per-agent tool allow-lists, Admin and Member roles (groups in Pro), and an HMAC-signed, hash-chained audit trail. AnythingLLM's agents can act too, but its Gmail, Outlook, Telegram and scheduled-job features work only in single-user mode. If you want governed agents that take actions in business systems like Odoo and email for a whole team, and a verifiable record of what they did, Pinchy is the closer fit.
AnythingLLM's desktop app and Docker self-hosting are free. Its hosted cloud plans are Basic at USD 50 per month and Pro at USD 99 per month, with Enterprise (on-premise deployment, SSO, RBAC) on quote, as of 16 October 2026. Pinchy's Community edition is free under AGPL-3.0 with no user limit; Pro is EUR 1,188 per year (EUR 99 per month billed annually) for up to 10 users, with a quote above that. Neither Pinchy license covers your servers or your model usage.
AnythingLLM has multi-user roles (Admin, Manager, Default) in its Docker version and event logs that record logins, messages sent, settings changes and document uploads. Agent skills are switched on for the whole instance, and no signing or chaining of the log is documented. Pinchy gives each agent a tool allow-list that starts empty (default-deny), and records every tool call in an audit row that is HMAC-SHA256-signed and hash-chained, with a built-in verifier and CSV export. Note that Pinchy's granular custom RBAC and SSO are on the roadmap, not shipped yet.
Choose AnythingLLM when your main goal is dead-simple local document Q&A, you want a polished desktop app for a single user or a small setup, and you value the no-code agent flow builder, MCP support and broad LLM-provider support. It also fits a personal assistant on your own Gmail or Outlook with scheduled jobs and a Telegram bot, in single-user mode. Choose Pinchy when you need governed agents for a team: per-agent permissions, roles (groups in Pro), a signed, hash-chained audit trail, and agents that act inside business systems like Odoo and email in a multi-user instance. Pinchy means running a Docker Compose stack and choosing a model; there is no desktop app.
Self-host Pinchy yourself in minutes, or book a call to talk it through. Your choice.
Or email us: info@heypinchy.com