Model or dataset
relaticle/relaticle avatar
relaticle/relaticle

relaticle says 37 tools on its repository page and 39 in its README

Open-source CRM with native AI agent support. 37 MCP tools, REST API, self-hosted. Built with Laravel & Filament

1,845 stars233 forksPHPAGPL-3.0

At a glance

What is it?
A self-hosted CRM with an MCP server for AI agents, built on a current Laravel stack. The configuration files are where the real detail is: two variables the container refuses to start without, defaults written as the literal string null, and a commented warning in the environment template describing an invite flow that silently loses the workspace.
Who is it for?
Relaticle is a credible self-hosted CRM if you want agent access to your own data without handing it to a vendor, and the AGPL licence with a self-hosted posture is a deliberate choice rather than an accident. Three things to check before you commit.
Can I use it commercially?
Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
Is it still maintained?
Yes. The repository received new commits within the last day.
What is it written in?
Mainly PHP, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • relaticle/relaticle1,845 ★ · +201
Star history
Repository2026-09-112026-11-15
relaticle/relaticle1.6k1.8k

201 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

The repository page and the README disagree on the tool count

The repository description says the project ships thirty seven MCP tools.

The README says thirty nine, and it says it twice: once in the opening paragraph describing what an agent connects to, and once in the list of core strengths describing the agent-native infrastructure.

So the number a visitor sees on the repository page is not the number in the documentation, and the documentation is internally consistent with itself. Whichever is right, one of the two published surfaces is stale, and the error sits on the field that a prospective user is most likely to evaluate the project on.

It is worth reading alongside the other MCP setting, because those two together are the security-relevant part of the agent surface: how many operations are exposed, and which origins may reach them. The origin list takes comma separated exact origins and the comment says wildcards are not supported, which is a stricter stance than most agent servers take.

That pairing makes the count discrepancy more than a typo. It is the difference between what you are told is exposed and what is actually reachable.

Two variables are required and everything else has a default

The compose file makes a deliberate distinction that is easy to miss.

Two values use a fail if unset syntax rather than a default. The application key has no fallback, and the database password has no fallback. If either is missing from the environment, the container refuses to start rather than booting with a placeholder.

That is the correct treatment for both. A default application key would mean every default install shares a signing secret, and a default database password would mean the same.

Everything else does carry a default, and they are sensible ones for a self-hosted deployment. Debug is off and the environment defaults to production. Logging goes to standard error at a warning level. The cache, session and queue stores all point at Redis, so a single additional service carries all three. Email defaults to writing to the log, which pairs with email verification being required by default, meaning a new account's verification link lands in the container logs rather than in a real inbox.

The published port defaults to 80 on the host while the container listens on 8080, so a quick start puts the application on port 80 without the operator choosing it. The install path from source is a clone and a single script:

bash
git clone https://github.com/Relaticle/relaticle.git
cd relaticle && composer app-install

Two defaults are the literal string null

Several values in the compose file default to a four letter word rather than to nothing.

The Redis password defaults to that word, and so does the session connection for the administrator panel. Read literally, the compose file appears to configure Redis with a password, and to point the admin session at a connection named by that same word.

This is the framework's own convention rather than a mistake. The application environment reader treats the string as an absent value, which is how a compose file expresses null for a variable that a PHP configuration layer would otherwise read as null. It is a widely used idiom in this ecosystem.

It is still a trap for anyone who is not expecting it. If you set a Redis password by editing the environment file, the change is trivially visible, but if you later inspect the running configuration expecting an empty password you will find a string instead, and if you ever change the application's environment reader or its case handling, values that were silently absent become the literal password. The same applies to the admin session connection.

Worth knowing before you debug either one.

The environment template documents a silent invite failure

The most valuable content in the configuration is a long comment warning about a specific failure, and it describes a bug rather than a setting.

The panel can be served from a path by default, or from a subdomain if you set one. The comment explains that switching to a subdomain requires a second change, because the session domain has to be set to the shared parent.

It then explains what happens if you do not. A guest opening a workspace invitation or a join link is redirected to the panel host to register, and the intended destination is remembered in the session. With the default host-only cookie, that session is not readable on the panel host. The new account therefore lands on the dashboard having silently never joined the workspace.

That is the shape of bug worth reading a comment about. There is no error, no failure banner and no failed request; the account is created and simply is not in the workspace, and the person inviting them sees nothing wrong.

The fix is one additional variable, which is exactly why it gets missed. Anyone moving from path to subdomain routing should read that comment before the first invitation goes out.

The image installs dependencies ignoring platform requirements

The container build has three stages and two of them contain a decision worth understanding.

The dependency stage installs from a lock file with development packages, scripts, autoloader generation and interactive prompts all disabled, preferring distributed archives. It also passes a flag that ignores platform requirements.

That flag is necessary rather than careless. The PHP extensions the application needs are installed in a later stage on a different base image, so at dependency-resolution time they are not present and the requirements could not be satisfied. The trade-off is that Composer will happily install a package whose runtime requirements the final image does not meet, and that failure surfaces at runtime rather than at build time.

The frontend stage contains a sharper comment. The package manager is installed by reading the version field out of the project's own manifest and installing that exact version globally, and the comment explains why: the bundled version-fetching tool was removed from a recent Node release, and letting the package manager fetch its own version needs a different C library than the Alpine base provides.

That is a precise, well documented workaround, and it pins the toolchain rather than floating it.

Five assistant config files and a server manifest

The root listing is unusually crowded with files that belong to other tools rather than to the application.

There are configuration directories and instruction files for several different coding assistants, and beside them a scatter of small JSON files whose purposes are not described in the README. There is also a custom git hooks directory, a file for including shared content across worktrees, a package store directory, and separate environment files for continuous integration and for testing.

Two of those deserve a name rather than a shrug. The server manifest is the standard file an agent client reads to discover a server, which is how the CRM gets into an assistant's tool list without anyone copying configuration by hand. The worktree include file is consistent with a project that has agents spawning isolated checkouts.

The package store directory is the one to look at. Stores like that are normally excluded from version control, and there is both an ignore file and a docker ignore file at the root, so whether it is tracked is something the directory listing alone does not settle.

Taken together this is a repository optimised for agent workflows, and the configuration surface is where you can see it.

Translations are a fork, and the requirements are unusually recent

Localisation is handled by a policy rather than by a contribution process.

The main repository ships English only. The codebase is described as ready for translation, and the instruction to a self-hoster is to drop a directory for their locale into place, which translates the entire application. A separate document holds the override workflow and what are called guardrails in continuous integration.

So a translation is not a pull request to upstream. It is a fork with a directory added, and the guardrails exist to keep that override from drifting against upstream changes. For a project whose pitch includes privacy and vendor independence, deferring localisation to forks is a coherent trade, and it does mean every translation tracks a moving upstream.

The other thing worth planning is the version floor, which is tight on every axis: a recent PHP, a recent PostgreSQL, a recent Node, a recent Laravel, and a recent release of the admin panel library. Two separate test configurations exist for normal and continuous integration runs, and static analysis, a formatter and an automated refactoring tool are all configured at the root.

The releases move on a patch-numbering scheme with three releases in the most recent month, and the most recent push is dated the day before this was written.

Editorial conclusion

Relaticle is a credible self-hosted CRM if you want agent access to your own data without handing it to a vendor, and the AGPL licence with a self-hosted posture is a deliberate choice rather than an accident. Three things to check before you commit. The tool count is stated two different ways across the project's own surfaces, so verify it from the server rather than the blurb. The allowed-origins list for the agent endpoint takes exact origins and refuses wildcards, so plan your deployment hostname before you need it. And read the environment template's subdomain warning, because the failure it describes is silent and the fix is a second variable that is easy to miss.

Frequently asked questions

What is Relaticle?

It is a self-hosted CRM with a REST API and an MCP server for AI agents, built on Laravel and Filament. It offers customizable field types including entity relationships, conditional visibility and per-field encryption without migrations, and multi-workspace isolation through five-layer authorization. It is licensed AGPL-3.0.

Does Relaticle lock you in to a particular AI vendor?

No. The MCP server is described as connecting any agent, naming Claude, GPT and open-source models, and the REST API is a separate surface with full CRUD. The project is self-hosted under AGPL-3.0 with your data staying on your own server.

How many MCP tools does Relaticle expose?

The README states thirty nine tools in two places, while the repository description states thirty seven, so the project's own surfaces disagree. Separately, the allowed-origins setting for the agent endpoint takes exact origins and does not support wildcards.

What does Relaticle need to run?

A recent PHP, a recent PostgreSQL, Composer 2 and a recent Node, with Redis for queues marked optional during development. The compose file refuses to start without an application key and a database password, and defaults email delivery to the log.

Can I serve the Relaticle panel from a subdomain?

Yes, by setting the panel domain variable, but you must also set the session domain to the shared parent. The environment template warns that without it, a guest registering from a workspace invite lands on the dashboard having silently never joined the workspace.

Official sources

  1. License: AGPL-3.0
  2. Project website
  3. README
  4. relaticle/relaticle on GitHub
  5. Releases
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/relaticle-relaticle.svg)](https://hysenlabs.com/projects/relaticle-relaticle)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/relaticle-relaticle.svg)](https://hysenlabs.com/projects/relaticle-relaticle)

Was this useful?