Every screen, explained.

Where you start, and the numbers at a glance.

Home

Home is the page you land on after signing in, and where you return when you activate a product. It shows a greeting, Quick Launch cards, your agent team and either a setup prompt or your recent projects.

• Quick Launch cards change with where you are: setup and a new product at first, then a new project with two starter templates, then your active projects, the dashboard, tasks and a Look Up card. A card with a /giljo badge can also be done from your coding tool.
• Starter templates appear until a product has its first project: Bootstrap a new product scaffolds folders and a README and proposes follow-up projects; Import an existing product analyzes an existing codebase and seeds 360 Memory.
• Your team lists the orchestrator and your active agents, with a slot count such as 3 / 16 slots: one reserved orchestrator plus up to 15 of your own. Manage opens the agent templates.
• Banners at the top give one suggestion each, such as enabling Git or tuning your context, with a Go to button. Dismissing one sticks across devices.

The welcome tour can be reopened any time from Tools > Startup.

Dashboard

The Dashboard shows statistics across your account: projects by status, by type and agents by role; counters for active projects, tasks, API calls, MCP calls and commits; a recent projects panel; and recent 360 Memory entries and Git commits. Click a project's serial badge to open its Project Review.

Define what you are building, once.

A product sits at the top of everything: Product, then Projects, then Jobs, then Agents. Every project, task, agent and memory belongs to one product. You can have several products and show several at once, each in its own tab.

Creating a product

1. Click New Product and enter a Name.
2. Attach one or more vision documents (.md or .txt, up to 5 MB each). You can add instructions to steer what the analysis pulls out.
3. The button moves through Stage analysis, Analyzing and Next while your AI coding tool analyzes the document and fills in the product fields. The other tabs unlock when it finishes.

To type everything yourself, tick Skip. A connected agent can also create the product and write its vision document from a single prompt.

Context fields

• Product Info: description, core features, brand and design guidelines, codebase folder.
• Tech Stack: languages, frontend and backend frameworks, databases, infrastructure, target platforms.
• Architecture: primary pattern, design patterns, API style, notes, coding conventions.
• Testing: quality standards, strategy, coverage target, frameworks.

Vision documents

A product proposal, architecture spec or feature brief is the most useful input you can give. It tells every agent the scope, constraints and intent, not just field values. No document yet? Start with one page: what the product does, who it is for, the core technical decisions and what success looks like. When a document changes, editing the product offers a one-click project to re-analyze it.

Tuning

When the stored context drifts from the real code, click the tune icon on the product card, pick the sections to refresh and click Generate Tuning Prompt. Paste the prompt into your coding tool: the agent compares the context with the codebase, shows its findings section by section, and applies only the changes you approve.

Create, stage and run a unit of work.

Creating a project

Click New Project. Give it a Name and a Description of what it delivers, and a Project Type such as BE, FE or API. The type sets the serial badge, for example BE-0001, and the next number fills in by itself, counted per product. New projects start inactive. Your coding tool can also create one with /giljo.

The projects list

• Click a project's colored serial badge to open its Project Review.
• Row actions include Activate, Deactivate, Complete, Cancel, Reopen, Review, Mark Superseded, Edit, Duplicate, Archive and Delete.
• Deleted projects stay in Deleted projects for 10 days, where you can restore them.
• Completed and Cancelled projects are locked. To redo the work, duplicate the project. Superseded keeps an old project visible with a link to its replacement.

The Jobs board: Staging and Implementation

Activating a project puts its card on the Jobs board, the one place a project is staged and run. The Staging side holds projects getting ready; the Implementation side holds projects that are running. A Staging card shows the project description, the mission once it is written, and the crew the orchestrator chose; you can edit a worker's mission before launch.

Execution mode

• Multi-terminal: you launch each agent in its own terminal. Phases run one after another; agents within a phase run in parallel.
• Subagent: one main agent starts and manages the others itself, in tools that support subagents.

Nothing is chosen for you. A read-only detected chip shows the tool Giljo HQ recognized. Running agents unattended is a separate account setting, and the dashboard follows a headless run live.

Staging

The stage button moves through Stage Project, Staging, Unstage to back out, and Re-Stage. Staging copies the orchestrator prompt, built from your product context, 360 Memory, project description and agent templates. When staging completes, Implement copies the implementation prompt and moves the card to the Implementation side. If the orchestrator finishes staging without choosing any agents, the project stays re-stageable.

Phases

Implementation: agents carry out the plan. Closeout: when every agent has finished, you review the project; Giljo HQ writes a 360 Memory entry and marks the project complete. That memory flows into the next project.

One ranked plan of what to build next.

The Roadmap ranks the current product's inactive projects and pending tasks, with a risk and complexity score on each. There is one roadmap per product.

Your AI agent builds it; Giljo HQ's server never analyzes or reorders it on its own.

1. Click Create Roadmap or Refresh Roadmap. This copies a prompt. Tick Add my own instructions first to steer it.
2. Paste the prompt into your connected agent.
3. A Waiting for your agent banner shows until the agent saves the roadmap.

Each card shows a PROJECT or TASK badge, its status, risk (low, med, high), complexity (light, med, heavy) and any reason the agent flagged it blocked. Drag the grip to reorder; the order saves by itself. Convert to Project promotes a task; Remove from roadmap takes an item off the roadmap without touching it. Fold in tasks shows projects alone. To run a project, activate it from Projects.

Watch every agent, decide, and close out.

The Jobs board has one card per project in flight. The All tab shows every product; a product tab narrows it. Compact view folds each card to one line, Detailed opens them all, and the board updates live without reloading.

Agent rows

A running card has one row per agent, orchestrator first: a play button that copies the agent's launch prompt (and re-copies it later), the agent badge, steps done out of total, elapsed time, status and unread messages. Jobs detail opens the full view with the message composer.

Statuses

Two agents of the same role are numbered for you: implementer, implementer-2, and so on.

Approvals

When an agent needs a decision, the card says Your decision needed and Jobs detail shows a Needs decision banner. Read the agent's reasoning in your AI chat, click the banner, and pick an option such as Approve, Reject or Defer. The dashboard is the one place a decision can always be made. Then tell the orchestrator to read its message and continue. Clients that support it may ask you to decide right in the chat instead, with the same result.

Closing out

1. When every agent has finished, Saving project memory shows while the 360 Memory entry is written.
2. Click Review project to read the summary: key outcomes, decisions and Git commits.
3. Click Close. The project leaves the board; reopen its summary any time from its serial badge.

How often waiting agents check in is one account setting, Agent Check-in Cadence, under Tools > Agents > Behaviour.

How a project was run, after the fact.

Click a project's serial badge on the Dashboard or Projects page to open Project Review. Its sections expand to show:

• Agent Jobs: each agent's mission, to-do list and progress.
• Agent Messages: the traffic between agents, useful for understanding why decisions were made.
• Project Threads: a read-only view of the project's Message Hub threads, with Open in Hub.
• Git Commits: commits recorded during the project, when Git integration is on.
• 360 Memory: the entry written at closeout.

Talk with your agents.

The Message Hub is where you and your agents talk. Its menu item shows unread messages and pulses when it is your turn somewhere.

• Threads sit under two tabs: Project threads, attached to a project, and General threads. Each row shows a CHT-#### id (click to copy), a status, the subject and the time. A search box narrows an open thread's messages.
• Replying: choose Broadcast for everyone on the thread or Direct for one agent, then press Ctrl+Enter or click Send.
• The baton shows only when it points at you: a Your turn badge above the composer and a gold chip on the thread. Replying passes the turn back. A directed message that needs action hands the turn to whoever it is addressed to.
• New Thread needs a subject; a project and product are optional. Only general threads can be deleted, and deleted threads can be restored.

Notes for later, and records of where work stopped.

One board holds two kinds of note. Tasks (TSK-) track work you have found but not scheduled. Handovers (HND-) record where a session stopped, for whoever picks it up next. Both share one serial counter.

Handovers

The reader of a handover cannot ask its author anything, so the board will not save one without three sections:

Ask your coding tool to hand the session over and it writes one for you. A handover is a record, so it is never converted to a project, and one still marked Pending cannot be archived.

Fields

• Categories: general, feature, bug, improvement, docs, testing.
• Priorities: low, medium, high, critical.
• Statuses: pending, in progress, on hold, completed, blocked, cancelled. Use On hold to park a task you have not decided on.

Filter by status and priority, or search by title, description or serial. Convert to Project turns a task into a new inactive project. Archived items leave the default view and come back through search or Show archived. Your coding tool can add and look up tasks with /giljo.

Connections, agents and context settings.

Tools in the left navigation has five tabs: Connect, Agents, Context, Notifications and Startup.

Connect

A directory of your tools with a live status for each: Connected, Waiting or Not set up. + Add a tool picks from Claude Code, Codex CLI, OpenCode and Generic MCP client and walks through the same connect flow as the Setup Wizard. The commands are in Connect your tools. The Git integration card lets Giljo HQ record commit history into 360 Memory from any local Git repository.

API keys

Click Generate API Key on a tool's card to create one. You can hold any number of keys, and each expires after 90 days. To retire one, use Revoke and confirm by typing DELETE. You get a notification a week before a key expires.

Agent templates

• Each product has 16 active slots: one for the orchestrator and 15 for your own agents. Slots count roles, not agents, so several agents with the same role share one slot. You can create as many templates as you like.
• Add Default Agents re-imports the starter set into the current product without touching your edits.
• A new product arrives with its crew switched on; an agent you create by hand starts switched off. Active here turns an agent on for the current product.
• A template has a role, an optional suffix, a coding tool, model and effort hints, a description, and Role & Expertise, the field that shapes how the agent behaves.
• There is nothing to install or sync: each agent receives its profile from the server when it starts work. Download profile (.md) gives you one to run yourself.

giljo_setup

Ask your coding tool to call the giljo_setup tool once after connecting, and again when a skills out of date banner appears. It installs the /giljo command and writes the Giljo HQ block into your project's CLAUDE.md or AGENTS.md. It does not install agents. If its download link goes stale, run it again.

Context settings

• Always on: product info and the project description.
• Toggles: tech stack, architecture, testing.
• Depth: vision documents at 33%, 66% or 100%; 360 Memory from the last 1, 3, 5 or 10 projects; Git history from 5 to 100 commits when Git integration is on.

Agent Silence Threshold and Agent Check-in Cadence, under Agents > Behaviour, accept 1 to 1440 minutes and default to 10. An admin can also override the orchestrator's core instructions, per product or for all products, with a Restore Default button; most people never need to.

What every project leaves for the next.

At closeout, agents write a 360 Memory entry: what was built, decisions made, patterns found and outcomes. Each following project starts with that history.

The Memory page searches every entry's summaries, outcomes and decisions. Filter by tag or project, group by project, and sort by newest, oldest or project order. Your agents can search the same history mid-run.

When an agent finds work it cannot finish in the current project, it files a follow-up task or project and cites it at closeout, so it lands on your Tasks board.

Your profile, sign-in and your data.

• Profile: username (fixed), first and last name, and email. On Hosted, a new email address is confirmed by a link sent to it.
• Password: at least 8 characters. Changing it signs you out everywhere.
• Forgot your password? On Hosted, Reset Password sends a link by email. On the Community Edition, a 4-digit Recovery PIN resets it from the sign-in screen.
• Google and GitHub sign-in on Hosted, managed under Account > Connected Accounts.
• Sign-in protection: 10 failed attempts lock the account for 15 minutes.
• Changing your email, password or PIN, or generating an API key, needs a live browser session; an API key alone cannot do it.

Your data

Download my data, under Account > Danger Zone, exports a ZIP of your products, projects, vision documents, agents, memory, tasks and configuration, with credentials removed. The download link is valid for 15 minutes. On Hosted, the Danger Zone also lets you download your latest backup, save a restore point, and request a restore. On the Community Edition your data lives in your own PostgreSQL database, so back it up like any database.

Trash

Deleted projects can be restored for 10 days; deleted threads, vision documents and agent templates for 30 days. Archiving only hides an item; you can unarchive it any time.

The numbers you are most likely to meet.