1 · Declined
Declined before starting: status stays, the task returns to its assigner, nothing is spent.
AI controls
Agents act. You stay in control.
What an agent may do, what it may only propose and what it cannot reach comes from the action registry, not from a policy document.
- Autonomy levels
- AI Inbox
- Audit log
- Undo
- Task
A person writes the brief and assigns it.
Fix the login bugperson - Work
Keeps the task and timer current. May set In progress or In review, never Done.
Updates the task and timeragent - Result
Posts what it found and attaches the work.
Adds a link to the pull requestagent - Approval
Proposes only. A person decides in the AI Inbox.
Ask to publish to the test siteagent proposes · person approves - Hard stop
Human only. These actions do not exist for agents. Attempts are logged.
Deploy to productionhard stop
Code teams: CI keeps running where it runs today. AlianHub reads the result from GitHub or GitLab and never builds or deploys.
A person checks before DoneNo agent closes a task in any project. A person closes it, whatever the project has chosen.
Pause every connected agentOne switch for the whole workspace. A change a person approves is still made.
Agents as teammates
Assign work to an agent like you would to a person.
Agents are members with an owner, a run budget and a scope. An AGENT tag marks them everywhere.
MemberRoleAccess
AAlex MorganOwnerEverything
SSam LeeMemberTheir projects
FFront Desk Request Sorter AGENTOwned by Alex Morgan · Suggests changesNo project scope · 1 skill
No seat. An owner, a run budget and a role’s permissions, minus anything destructive.
How an assignment ends
2 · Handed back
Partway, then a judgement call: work so far is posted, hours logged, the task moves to a person.
3 · Going wrong
Too long or off track: stopping pauses the agent, keeps its work and shows the cost.
The rule: an assignment can always return to a person, with the work preserved.
Route many tasks at once
Select tasks. The router proposes an agent per task and refuses those that need a person.
Team packs
Turn on a team’s role agents in one click. All start paused and the pack can be undone.
IT companyManufacturingAgencyE-commerceConstructionProfessional servicesEducationClinic administration
Small · 5 to 25Medium · 25 to 150Large · 150 to 1,000
A seat is one connected AI working a role. Plan one person to supervise 3 to 5 seats.
Connect your AI
Use the AI app you already use.
Connect Claude or ChatGPT once. It sees and changes only what you can.
ClaudeChatGPTClaude Code & CLI tools
AlianHubMCP
Claude & ChatGPT: connect by address, once your server admin allows itClaude Code & others: create a token, run the command. Shown once
11 READ AND WRITE TOOLSDoneDeleteMergeDeployAbsent from the tool list, not merely disabled.
What your AI can do here
On
Find and read tasks, add tasks and subtasks, comment, change status.
Server switch
Read projects, lists, statuses, comments, timesheets. Log time.
Server switch
Change tasks (assign, move, set fields, archive) and write docs.
Server switch
Tags, task links, lists, doc comments and goals.
Extra tiers are switched on by whoever runs the server. Agents that work round the clock need a server key.
Connections One screen with tabs for Apps, MCP servers and External agents
Apps and MCP servers
Any MCP server by URLExternal agents
Claude Code, Codex and other CLI agents use the Agent API from a personal account, scoped to chosen projects. Writes land in the AI Inbox first.
Where the agent runs Three ways, chosen by owners and admins
Workspace key
One API key in settingsPAYSThe company, capped per agent
RUNSServer-side, from AlianHub
CAN DOEverything on the allow-list, also scheduled
BEST FORScheduled agents, such as a daily digest
Personal AI account
Their own Claude Code. No key heldPAYSTheir own subscription. $0 workspace spend
RUNSOn their machine
CAN DOOnly what they can, while at the keyboard
OFFBOARDINGRevoke the token and the link dies
BEST FORHands-on work, on their own terms
Self-hosted model
Ollama or vLLM on your GPUPAYSNobody per run. Power and one GPU
RUNSInside your network. Nothing leaves
CAN DOThe allow-list, slower on hard reasoning
BEST FORRegulated teams, air-gapped installs
Observe
See what your AI did, what it cost and how well it worked.
Health, quality, spend and an audit log. No names on ratings.
AI health
1 hour24 hours7 days30 daysPer agentPer modelAlert thresholds
AI quality
7, 30 or 90 daysMost disliked, by consentHeld-out question setCost per feature
Audit log
AllAgents onlyOutside agentsGated actionsUndoneRefusalsSearchExport to CSV
Usage & spend
Rate limits per agentMonthly spend capsPause in one click
GET STARTED
Agents act. You stay in control.
$ curl -O https://raw.githubusercontent.com/aliansoftwareteam/AlianHub-Project-Management-System/main/docker-compose.yml$ echo "JWT_SECRET=$(openssl rand -hex 32)" > .env$ docker compose up -d# AlianHub running at http://localhost:4000$ npm run setup✓ Opens localhost:4000/#/setup→ demo.alianhub.com✓ Demo workspace, nothing to installOpen source. Self-hosted. No per-seat pricing. Or open the live demo first.