Self-Hosted AI for Project Management: Setup Guide
11 hours agoPUBLISHED INAgile
|
Quick answer Self-hosted AI for project management means the AI service runs on your own servers, so your project data, vectorized content and settings stay in your network. In Sanplex, the Agent depends on a service called ZAI, which installs with Docker and a PostgreSQL database that supports vectors. The catch: where your prompts go still depends on which language model you connect, so decide that before you promise anyone data never leaves. |
Plenty of teams that like the idea of AI in their project tool hit a wall at the first security review. Where does the data go? Who sees the prompts? Can it run where our other systems run?
Self-hosting is the honest answer for some of them. This guide explains what self-hosted AI for project management involves, what Sanplex's documentation describes for setting it up, and the trade offs, including the one people tend to miss. We'll say where Sanplex fits, and where it doesn't.
What is self-hosted AI for project management?
It's AI features that run on infrastructure you control, instead of a vendor's cloud. The AI service, the data it uses and its settings live on your servers. For project work, that usually means an assistant that can read your stories, bugs and documents without those leaving your network.
-
Your servers. You install and run the AI service yourself.
-
Your data. Vectorized project content is stored in your own database.
-
Your settings. Models, keys and integrations are managed by your admin.
-
Your upkeep. Updates, backups and monitoring are your job.
-
One open question. Which language model answers the prompt, and where does it run?
What Sanplex's documentation describes
The Sanplex guides describe a setup with three parts. First, the Sanplex Agent plugin, installed from the admin area. Second, a service called ZAI, which the Agent depends on. Third, a connection between them, with an integration ID and secrets that your admin keeps safe.
|
Part |
What it does |
Read more |
|---|---|---|
|
Sanplex Agent plugin |
Adds the Agent to the Sanplex interface once it's connected |
|
|
ZAI service |
The underlying AI service. Installed with Docker |
|
|
ZAI admin panel |
Model setup, integrations, and pricing rules for model usage |
|
|
Data vectorization |
Turns Sanplex data into a searchable form for chat |
What you need before you start
-
A server for ZAI. With Docker and Docker Compose installed.
-
A PostgreSQL database. The guide describes one with vector support, either from the included Compose setup or one you manage yourself.
-
Network access. Between Sanplex and the ZAI service, on the port you choose. The default is reported as 8000.
-
HTTPS plan. If Sanplex uses HTTPS, ZAI should too, usually through a reverse proxy and certificates.
-
An admin. Someone to own installation, secrets and updates.
-
A model decision. Which language model or provider ZAI will use.
The setup in stages
|
Stage |
What happens |
Common trap |
|---|---|---|
|
1. Prepare the server |
Install Docker and create a folder for ZAI |
Skipping a plan for backups |
|
2. Create the Compose file |
Define the ZAI app and its database |
Copying settings without understanding them |
|
3. Start the service |
Pull images and start the containers, then check the logs |
Not reading the logs when something fails |
|
4. Initialize ZAI |
Open the service in a browser and complete setup |
Leaving default or weak secrets |
|
5. Install the Agent plugin |
Install from the admin area, then refresh |
Skipping the plugin cleanup step in the guide |
|
6. Connect and test |
Enter the port, integration ID and secrets, then test a chat |
Browser blocks from content security rules |
|
7. Enable vectorization |
Let Sanplex data sync, then check a question |
Forgetting it reflects the last synchronization |
Two details from the guides catch people out. A fresh install is required if you used an early beta of ZAI, because ZAI 1.0 doesn't upgrade in place from those. And if the browser blocks the connection, the guide describes adding the ZAI address to Sanplex's content security rules in its configuration file. Read the live guides for the exact steps.
How the Docker part works, in plain words
Docker runs the ZAI service in a container, a sealed package with everything it needs. A Compose file describes the pieces: the ZAI app and, optionally, a PostgreSQL database with vector support. You start the pieces with a couple of commands, check the logs to see that they're healthy, and open the service in a browser to finish setup.
The upside is repeatability. The same file can recreate the service on another server. The downside is that you're now running a service, with the usual chores: updates, monitoring, backups and keeping secrets out of the wrong hands.
The trade off people miss: where do prompts go?
Self-hosting ZAI keeps the service and the vectorized data on your servers. It doesn't automatically mean the language model is local. The ZAI admin panel lets you configure models and describes an interface that can reach models from several providers. If you connect an outside provider, prompts and the project context included with them travel to that provider.
|
Setup |
Where the service runs |
Where the model runs |
What to confirm |
|---|---|---|---|
|
ZAI on your server, outside model provider |
Your servers |
The provider's cloud |
What the provider does with prompts |
|
ZAI on your server, model you host |
Your servers |
Your servers |
Hardware, model quality and upkeep |
|
Vendor cloud AI |
Vendor |
Vendor or provider |
Data terms and location |
If your requirement is that nothing leaves the building, the model choice is the deciding factor. Write that into your security review early.
Security and governance basics
-
Secrets. Store the integration ID and secrets safely, and rotate them if shared.
-
HTTPS. Match the protocol Sanplex uses.
-
Access to the admin panel. Limit who can change models and integrations.
-
Permissions in chat. The Agent guides note that data vectorization doesn't currently enforce Sanplex permissions, so decide what to vectorize.
-
Backups and updates. Plan both before go live.
How to measure success
-
Security sign off. Did the setup pass your review?
-
Uptime. Is the service reliable enough for daily use?
-
Answer quality. Test twenty questions against the source.
-
Admin hours. Time spent on upkeep each month.
-
Adoption. Are people actually using it?
Self-host, cloud or skip?
|
Approach |
Good for |
Watch out for |
|---|---|---|
|
Self-hosted AI service |
Teams with strict data rules and an admin |
Setup, upkeep and the model question |
|
Vendor cloud AI |
Teams that want zero upkeep |
Data leaves your network |
|
Build your own |
Teams with engineers and special needs |
Everything is yours, including bugs |
|
No AI for now |
Teams not ready for the governance work |
Missing time savings |
How Sanplex fits
Sanplex is a full lifecycle project management tool and a Jira alternative for Agile, Waterfall and CMMI teams, built by EasySoft INC. Scrum, Kanban, stories, tasks, bugs, test case management, documents, product planning and reports all live in the one tool.
Sanplex can run on your own servers. The on-premises option can sit on an internal network isolated from the internet, and the guides describe a self-hosted ZAI service for the AI side. Sanplex Agent, the AI assistant, writes story descriptions, creates bug reports from keywords and turns stories into test cases. It's available in the Standard and Premium plans.
On price, Sanplex Free 4.0 is free forever for unlimited users. Standard is $49.9 per user per year and Premium is $69.9 per user per year. The pricing page has the details, and plans change, so check the pricing page before you decide.
Where Sanplex does not fit
Self-hosted AI takes real admin effort, and the Sanplex guides describe a multi part setup. If you want AI with no infrastructure work, a cloud setup will be simpler. The guides also say data vectorization doesn't currently enforce Sanplex permissions, which matters for strict environments. And in a fully isolated network, you'd need a model you can reach from inside it, so check that with the guides before committing. Agent features sit in the paid plans, not Sanplex Free 4.0.
A worked example
This is an illustration, not a case study. A 40 person engineering group in a regulated company wants AI help in their project tool without sending project data outside.
-
Requirements. AI chat over stories and bugs, data kept in the company network, security sign off before launch.
-
Build. An admin installs Docker and ZAI on an internal server, connects it to Sanplex and decides which model it will use.
-
Test. They run twenty real questions on a non sensitive project and check each answer, and the security team reviews where prompts go.
-
Deploy. They enable the feature for one team, with a note on what the chat can and can't see.
-
Maintain. The admin schedules updates and backups, and the team reviews answer quality every month.
Questions for your security team
-
Where does the model run? Our servers, a private cloud or a public provider.
-
What leaves the network? Prompts, project context, logs.
-
Who administers it? Named people, with access reviewed.
-
How are secrets stored? And who can read them.
-
What's the update plan? Who patches the service and how fast.
Having these answers written down before the pilot makes the sign off a short conversation instead of a long one.
Common mistakes to avoid
-
Assuming self-hosted means private. Check where the model runs.
-
Skipping the logs. They tell you what failed.
-
Weak secrets. Keep them safe.
-
Forgetting upgrades. Plan how you'll update the service.
-
Enabling everything at once. Start with one project.
Checklist before you start
-
A server with Docker and enough resources
-
A PostgreSQL plan, bundled or your own
-
HTTPS and reverse proxy decided
-
A model and provider chosen, with data terms read
-
An owner for secrets, backups and updates
-
A pilot project selected
-
The live Sanplex guides read end to end
Where to go next
Start by reading the guides, then try Sanplex on a test project with Sanplex Free 4.0. The AI features sit in the paid plans. The whole product is at sanplex.com.
Frequently asked questions
What is self-hosted AI for project management?
Self-hosted AI for project management means the AI service runs on servers you control, so project data and settings stay in your network. Which language model answers still needs a decision.
Can I run Sanplex Agent on my own servers?
The Sanplex guides describe installing a ZAI service with Docker, which the Agent depends on, and Sanplex itself can run on premises. Check the guides for how the pieces apply to your setup.
What does ZAI need to run?
According to the installation guide, Docker, a Compose file and a PostgreSQL database with vector support, which can be the bundled one or your own.
Does self-hosting mean my data never leaves my network?
Not automatically. The service and vectorized data stay on your servers, but prompts go wherever your configured language model runs. If you use an outside provider, they leave.
Is self-hosted AI cheaper than cloud AI?
Not necessarily. You avoid some vendor fees, but you take on server, upkeep and security work, plus model costs.
Can AI work in an air gapped network?
It can if the model is reachable from inside the isolated network. Confirm this with your vendor and security team.
What are the risks of self-hosting AI?
Misconfigured secrets, unpatched services, unclear data flow to model providers and permissions in chat. Review each before go live.
How much does Sanplex cost?
Sanplex Free 4.0 is free forever for unlimited users. Standard is $49.9 per user per year and Premium is $69.9 per user per year. Sanplex Agent is in Standard and Premium. Check the pricing page, since plans change.
Where do I find the setup steps?
In the Sanplex guides for ZAI service installation, Agent plugin installation and ZAI configuration. Use the live guides for the exact commands.
Keep reading
|
Need AI that stays inside your network? Read about running Sanplex on your own servers: explore on-premises. |
Resource
- Blog
- Customer stories
- FAQ
Support
- Book a Demo
- Email Us: [email protected]
2026-11-18 14:26:04
0