A chief of staff and a team of AI specialists

Your agents ship autonomously.

AidenHQ gives you a whole AI team that runs itself. Tell your chief of staff what you need, and specialists plan it, build it, ship it and follow up, around the clock, on your own server.

curl -fsSL https://get.aidenhq.ai | sh

One command on a Linux server with Docker and Node.

Not a chatbot. Not a workflow builder. A team that works.

A chatbot waits for your next prompt. A workflow breaks on the first case nobody drew. AidenHQ agents take a goal, work out the steps, use your tools and keep going until the job is done.

Hand over real work, in a sentence.

You want powerful agents but have no time to build them. Start with the jobs owners hand over first. No setup per task, no code.

A chief of staff that follows up for you

Aiden tracks every commitment made in Slack and email, chases people when things slip, and briefs you every morning on what moved.

An app from one sentence

Say "build me a roadmap tracker". Agents design it, build it and put it live for your team to sign in to, then keep it running.

A new agent from a description

Describe the agent you wish you had. The Agent Architect writes its job, picks its tools, runs it on trial and graduates it once its work holds up.

An engineering team that ships

Not one coding bot: a designer, engineers, a reviewer, DevOps and security, each a specialist. An issue becomes designed, built, reviewed and deployed work.

Stop babysitting your agents.

Most agents stall, loop or wait for the next perfect prompt. Here you talk to Aiden like you would a chief of staff. The work fans out to specialists who know their job, and Aiden chases it until it lands.

  1. 1

    Say what you need

    On Telegram, Slack or the web. Ask for an app, forward a thread, set a deadline.

    Chase the vendor contract and tell me by Friday.
  2. 2

    Aiden delegates

    Your words become tasks for the right specialist. Aiden chases them and keeps the thread with you.

    Follow-up TrackerVendor contract, due Fri
  3. 3

    Specialists do the work

    Research, code, deploys, follow-ups. They wake on schedules and events, day and night.

    08:00 Researcher wrote the weekly report
    08:14 Engineer opened PR #118
  4. 4

    Done means checked

    A grader checks the work before a task counts as done, and an AI reviewer checks every risky action. Then it ships.

    pr-tracker v1 is liveReviewed, merged, deployed

Apps that look after themselves.

Most AI-built apps are demos that rot after launch. Your agents ship dashboards, trackers and internal tools to your own apps platform, then watch them, patch them and fix them.

pr-trackerBuilt by Engineer, run by DevOps Engineer
Pull requestAuthorWaitingStatus
Retry on flaky uploaddana3hApproved
Billing export to CSVlee1dIn review
Cache warm-up on deploysam2dNeeds reviewer
Drop legacy auth pathkim4dCI failing
Dark mode for reportsana5hApproved
Rate limit the public APIraj1dIn review
Fix timezone in digestslee2hMerged
model-spendReads the shared warehouse
MonSun
  • Your team signs in. Colleagues use your company login and see only the apps you share.
  • Built on your data. Apps read the shared warehouse, each with its own private schema.
  • Sandboxed. No access to the server unless granted. Secrets arrive as files.
  • Never abandoned. Health checks, rollbacks and patches run on their own.

It keeps running, and learns your corrections.

Agents that forget repeat the same mistakes. Yours remember what worked, turn every correction into a rule, and keep their own work in shape after launch.

  1. Mon 09:12
    A deploy failed its health check.

    The DevOps Engineer rolled the app back to the last good version.

  2. Mon 11:40
    "Never chase people on weekends."

    Your correction became a rule in the Follow-up Tracker's playbook.

  3. Tue 03:00
    A vulnerable base image was found.

    Security Watch opened a patch PR. It passed CI and review, then merged.

  4. Wed 14:05
    A broken mobile layout never shipped.

    The UI Reviewer caught it in screenshots before the merge.

  5. Fri 17:00
    Weekly fleet review.

    Two repeat defects traced to one playbook line. The Agent Coach proposed the fix.

After launch

  • Every app's health is watched, and a bad deploy rolls back on its own.
  • Incidents are raised and fixed.
  • Dependencies and security issues are patched with pull requests.
  • UI reviews catch visual regressions before they ship.
  • Every change goes through CI and AI review, so nobody has to babysit.

Every week

  • Every correction becomes a rule or an automated check, so the mistake stays fixed.
  • Agents keep a memory of what worked.
  • A coach agent reviews runs and proposes better playbooks.
  • Weekly fleet reviews track repeat defects and reverts, and fix their causes.
  • Agents propose changes to their own instructions and tools, tested before they land.

Agents that build agents.

Building an agent usually means prompt engineering and glue code. Here you describe the job in plain words. The Agent Architect designs the agent: its role, its playbook, its tools and when it wakes. It runs on trial, gets graded on real work, and joins the team when it holds up.

  • A playbook, not a prompt. Every agent has a written handbook you can read and edit.
  • Only the tools it needs. Access is granted tool by tool.
  • Reviewable. Each new agent arrives as a change in your git repository.
agents/support-watcher/agent.yamlOn trial
name: Support Watcher
reports_to: aiden
wakes_on: new message in #support
tools:
  slack: read
  tasks: write
model: ~deepseek/deepseek-flash-latest
playbook: AGENTS.md

Plugs into the tools you already use.

Agents that can't reach your tools can't do your work. Connect each system once. Every agent you allow can use it, tool by tool, and every source feeds one shared warehouse.

  • Slack and GmailFollow-ups that never drop
  • GitHub and JiraA live engineering status board
  • Grafana and ClickHouseIncident digests with the graphs attached
  • Calendar, Drive and GleanMeeting prep with every doc you need
  • TelegramYour chief of staff in your pocket
  • The warehouseOne database every app can build on

Add, swap or extend any capability.

Most agent products lock you into one model, one engine, one way of working. AidenHQ is built from modules: engines, channels, connections, agents and apps. Add one, swap one, remove one. Ours cover most needs, and the marketplace is open for anyone to extend.

  • Swap the engine. Engineers run on Codex, Claude Code or opencode.
  • Same rules for everyone. Third-party modules use exactly the public contracts ours do.
  • Held to their word. Each module declares what it can read and write, and is checked on every call.
  • CodexCoding engine
    Installed
  • Claude CodeCoding engine
    Available
  • Chief of staffAiden, your front door
    Installed
  • EngineeringEngineers on the code-gated runner
    Installed
  • TelegramTalk to Aiden from your phone
    Installed
  • SlackAiden in your workspace
    Available
  • Google WorkspaceGmail, Calendar, Drive
    Installed
  • JiraIssues and projects
    Available
  • MeetingsNotes, decisions, follow-ups
    Available
  • SecuritySecurity Watch and Red Teamer
    Available
  • opencodeCoding engine
    Available
  • GrafanaDashboards and alerts
    Available

Yours, end to end.

Your agents, your data and your know-how shouldn't live in someone else's cloud. AidenHQ runs on one server you control, on the models you choose, with everything in your own git repository.

  • Your server. One machine, reachable only on your private network.
  • Any model. Pick any OpenRouter model for each agent, and change it whenever you like.
  • Everything in git. Every agent, playbook and app is a file you can read, review and revert.
  • A small kernel. For the technical: a small core holds the gates, identity, sandbox, metering and audit log. Everything else is a module on the same public contracts.
agents.yaml
aiden:           model: ~deepseek/deepseek-flash-latest
architect:       model: ~deepseek/deepseek-flash-latest
ai-reviewer:     model: ~openai/gpt-luna-latest
engineer:        engine: engine-codex
acme-instance
$ git log --oneline
a41c9e2 feat(apps): add pr-tracker            Platform Engineer
7f03b18 fix(playbook): wait 2 days to chase   Agent Coach
5be1e77 feat(agents): add Support Watcher     Agent Architect
$ git revert 7f03b18
Reverted. The fleet picks it up on the next apply.

Set the cap. No surprise bills.

Autonomous agents can quietly run up a bill overnight. Here every agent has its own spend caps, runs on fast, efficient models by default, and every model call is on the ledger.

  • Caps per agent. Limits per run, per task and per day. An agent stops at its cap.
  • Efficient by default. Most agents run on a fast, efficient model. Premium models only where they earn it.
  • A full ledger. Every call is metered and logged by agent and task, with one key limit on top.
Spend todayCaps per agent
  • Aiden$6.60 of $30
  • Engineer$14.40 of $30
  • Agent Architect$9.30 of $30
  • Follow-up Tracker$2.70 of $30
  • DevOps Engineer$4.20 of $30

Agents decide. Code approves.

Agents that act unchecked are a risk no one should take. AidenHQ checks every action before it happens, attacks itself to find holes first, and locks down the rest. Every action is on the record, and you keep the final say on what matters.

Every action on the recordToday
  1. EngineerPush feat/pr-tracker to acme-agents/pr-trackerRan
  2. DevOpsDeploy pr-tracker v1.3 to the apps platformReviewer approved
  3. Follow-upSend a reminder to an allowlisted teammateReviewer approved
  4. EngineerPush a branch with a token in the diffBlocked
  5. PlatformRaise the Engineer day cap to $45Your call

Routine, verified writes run at once. An AI reviewer judges the rest against a written policy. Changes to spend, autonomy and what agents may write stay your call.

  • Ran at once
  • AI reviewer
  • Your call
  • Blocked

Every action checked

AI review of every write
A reviewer model checks each push, merge, deploy and message first.Changes that expand authority get a second reviewer from another model family. Every false reject becomes a test case.
Rules in code, below the model
No prompt can argue its way past a gate.Tiers come from verified state, never from text an agent wrote.
Per-tool permissions
Each agent gets exactly the tools it needs.Every tool call passes one proxy that checks the agent's grants.
Prompt-injection defence
Messages, emails and documents are data, never orders.Untrusted text is fenced before any agent reads it.

Attacked and patched

A Red Teamer agent
A dedicated agent attacks your own system to find holes first.It runs on a separate model family suited to offensive work.
Security Watch
Scans every image and dependency for known vulnerabilities, continuously.Fixes arrive as patch PRs through the same review.
Supply-chain discipline
Nothing unpinned or brand new gets in.Every dependency pinned exactly, a 7-day minimum age, gitleaks secret scanning and image scans.

Locked down

Sandboxed engineers
Coding agents work in a sealed box.Each job is jailed with its own network, no secrets and read-only tokens.
An encrypted vault
Secrets stay encrypted and out of every prompt.Each service gets only its own secrets, as files.
Private network only
No public ports. Shared apps sit behind your company login.Board on your LAN or Tailscale; Okta or other SSO for shared apps.
Spend caps and audit log
Every action and model call is on the record.Per-agent caps, plus a log of gate decisions and tool writes.

One line to install.

Run it on your Linux server. Aiden says hello on your board when it's ready, and your team is waiting for its first job.

curl -fsSL https://get.aidenhq.ai | sh