Pinchy vs amber
amber (formerly amberSearch) is a managed business AI platform: permission-aware search across your silos, a company GPT, and agents that call tools — operated by amber in T Cloud data centres in Germany and the EU. Pinchy is open source (AGPL) and self-hosted: you run it on your own servers, with a fail-closed allow-list per agent and a signed audit trail. Both take GDPR seriously. The real question is who operates the system, and which of your systems it can reach.
The Core Difference
Four modules in one hosted platform: amberSearch across your silos, amberAI as a company GPT with sourced answers, amberAgents for recurring tasks, amberProjects as a shared workspace. amber runs all of it, with a choice of hosted models and 70+ connectors. Nothing to install.
One Docker stack on your own servers, with your own models. The centre of gravity is governed action: which tools each agent may use, a signed record of every call, and the option to run fully air-gapped.
amber is closed source and operated by amber, with your data in its cloud until you reach the Enterprise tier. Pinchy is AGPL-3.0: you can read the code, change it, patch it on your schedule, and run it on infrastructure no vendor can reach.
Side by Side
Figures as of 26 August 2026, taken from amber's public pricing, product, integrations and security pages and its help centre. Check amber for current details.
| Pinchy | amber | |
|---|---|---|
| Category | Self-hosted agent platform | Managed business AI platform (search, company GPT, agents) |
| License | AGPL-3.0 (open source) | Proprietary (closed source) |
| Who operates it | You (Docker, your servers) | amber (T Cloud, Germany & EU) |
| Self-host / on-prem | Yes — the default, at any size | Enterprise tier (from 500 seats) |
| Where your data lives | Your infrastructure | amber's data centres (your own from Enterprise) |
| Entry point | Free under AGPL, from one user | Starter €9.99/user/mo, from one user |
| Paid pricing | Flat Pro tier, not per seat | €13.99–17.99/user/mo, from 50 users |
| Search across silos | Focused connectors | Core strength — 70+ connectors, from the Professional tier |
| Model choice | Any, including local models via Ollama | Hosted choice: OpenAI, Anthropic, Gemini, Mistral, Llama, open source |
| Reaching internal systems | Any host on your own network | MCP servers need a public domain name; localhost not supported |
| Who approves a tool call | An admin grants each tool per agent, fail-closed | Per tool, the user approves in the conversation ("always ask" / "always allow") |
| Agent actions (write / automate) | Native, scoped per agent | Native via MCP tools, read-only or write |
| Audit of agent actions | Per-row HMAC-SHA256 signature, built-in verifier, CSV export | Logged and attributable to user accounts (ISO 27001, SOC 2) |
| Fully offline / air-gapped | Yes (local model via Ollama) | Not in the managed cloud |
| ERP-native (Odoo) | Deep, first-class (read + write) | Via connectors and MCP |
| Channels | Web UI + per-agent Telegram bots | Web, Teams, desktop, mobile, Chrome, Outlook, embeddable widget |
| Data residency / compliance | GDPR by architecture (your servers) | GDPR + EU AI Act, ISO 27001, SOC 2, hosted in Germany & the EU |
Being Honest
amber's core is permission-aware retrieval across SharePoint, Confluence, Microsoft 365, Jira, DATEV and 70+ connectors, with source attribution and multilingual results. It inherits the access rights you already have rather than asking you to rebuild them. For company-wide "find and ask across everything", it is purpose-built — and amber reports 400+ customers and over 100,000 users.
amber runs it for you: live in hours, no data migration, no server to patch. Pinchy asks the opposite of you — a machine to keep running, updates to apply, a model to choose. That is real work, and amber takes all of it off your desk.
amber's Starter tier begins at a single user, so a two-person team can be on a managed German platform this afternoon. Two thresholds sit further up: search across your connected systems starts at the Professional tier with a 50-user minimum, and hosting on your own infrastructure is an Enterprise option from 500 users.
Platform and index run in T Cloud data centres in Germany and the EU, with ISO 27001 certification, SOC 2, and alignment to GDPR and the EU AI Act. For organisations that want a managed EU assurance rather than running their own stack, that is a simple, strong story.
Search, company GPT, agents and shared projects, plus a choice of hosted models from OpenAI, Anthropic, Gemini, Mistral and open source, deep research, image generation, and clients for Teams, desktop, mobile, Chrome and Outlook. A broad assistant the moment it is switched on.
Where Pinchy Wins
amber's help centre states that it supports MCP servers reachable under public domain names only, and excludes localhost for security reasons. So the tools a hosted platform can act through are the ones you expose to the internet. Pinchy runs inside your network: the ERP on a private subnet, the file share, the internal API — reachable without publishing any of them.
One Docker Compose stack on your own servers, your own models. Paired with Ollama, nothing leaves your network — air-gapped if you need it. That is a stronger guarantee than a hosting location, because there is no transfer to trust: the data never leaves in the first place.
In amber a tool is approved per tool, by the user, inside the conversation — "always ask" or "always allow". In Pinchy an agent has no tools at all until an admin grants each one, and file access stays scoped to directories the admin picked. Nobody widens their own agent's reach, and a platform update never hands an agent a capability you didn't choose.
amber logs agent actions and attributes them to user accounts. Pinchy signs every audit row with its own HMAC-SHA256 signature, so changing any field invalidates it, and a built-in verifier re-checks the whole log in one click. Export it as CSV into your SIEM. Tamper-evident, not tamper-proof — and it sits on your infrastructure, not your vendor's.
Pinchy is AGPL-3.0. Read the code, modify it, fork it, patch on your own schedule. With amber you depend on a proprietary vendor, and below the Enterprise tier your data sits in their cloud.
Pinchy ships a deep Odoo integration: query stock, create orders, reconcile invoices, all from chat or Telegram. Acting inside an ERP is a first-class use case, not a generic connector.
Pinchy is free under AGPL, and its paid tier is flat. Growing from five people to fifty changes what you run, not what you pay. amber is priced per user per month from €9.99, and the tiers that search your connected systems carry a 50-user minimum.
Decision Guide
You want someone else to operate it: company-wide search across many silos, live in hours, no infrastructure of your own, hosted in Germany and the EU. The systems it needs to reach are reachable from their cloud, and a managed contract is the assurance you want. It works from a single user upwards.
You want to run the platform yourself: your servers, your models, reaching internal systems that were never meant to be public. You want an admin-set allow-list per agent and an audit log you can verify and export. Free under AGPL at any team size, and especially if Odoo is central or you need an air gap.
It isn't how big you are — both work for a team of three. It's where the system should run, and which of your systems it has to reach. Reachable from a managed cloud, operated for you? That's amber. Inside your own network, operated by you? That's Pinchy.
FAQ
Partly — they overlap but differ in shape. amber (formerly amberSearch) is a managed business AI platform: permission-aware search across your knowledge silos, a company GPT, agents for recurring tasks, and a shared project workspace, all operated by amber in T Cloud data centres in Germany and the EU. Pinchy is a self-hosted, open-source agent platform you run on your own infrastructure, with a fail-closed allow-list per agent and a signed audit trail. Both are GDPR-strong. The difference is who operates the system and which of your systems it can reach.
No. amber's Starter tier begins at a single user, so small teams can use it. Two thresholds do sit further up the range: search across your connected systems starts at the Professional tier, which has a 50-user minimum, and hosting on your own infrastructure is an Enterprise option with a 500-user minimum. Pinchy has no such thresholds — it is free under AGPL and self-hosted from the first user.
At amber's Enterprise tier, which lists hosting on your own infrastructure and carries a 500-user minimum. Below that, amber is managed SaaS, hosted in T Cloud data centres in Germany and the EU and operated by amber. Pinchy is self-hosted from day one at any size: a single Docker Compose stack on your own servers, and it can run fully air-gapped with a local model via Ollama.
They act. amber connects tools through MCP servers, and its help centre describes tools as read-only or write, with the user approving them inside the conversation — "always ask" or "always allow". The difference from Pinchy is not whether agents act, but who authorises it and how far it reaches: in Pinchy an agent has no tools until an admin grants each one, file access stays scoped to admin-approved directories, and every call lands in a signed audit row. amber's help centre also states it supports MCP servers under public domain names only, excluding localhost — so internal systems have to be exposed before a hosted agent can act on them.
amber is GDPR-compliant by hosting your data in T Cloud data centres in Germany and the EU that it operates, with ISO 27001, SOC 2 and EU AI Act alignment. Pinchy is GDPR-compliant by architecture: it runs on infrastructure you control, so data never leaves your network in the first place — and paired with a local model via Ollama it can be fully air-gapped. amber gives you a managed EU cloud; Pinchy gives you your own servers.
No. amber is a proprietary, closed-source platform. Pinchy is AGPL-3.0: you can read the code, modify it, and run it yourself — no vendor holds your data or your deployment.
Choose amber when you want a managed platform someone else runs: company-wide search across many silos, live in hours, with German and EU hosting and no server of your own. Choose Pinchy when the system should run inside your network — because your ERP or file shares have no public address, because you want your own models or an air gap, or because you want an admin-set allow-list per agent and an audit log you can verify yourself. Company size is not the dividing line: both work for a team of three.
Self-host Pinchy yourself in minutes, or book a call to talk it through. Your choice.
Or email us: info@heypinchy.com