A team you hire once, with every role written down.
Roles, goals and tasks live in harnsy, not in one session’s head. Every agent that takes a role starts from its document.
Your team comes from a conversation.
Tell an agent how you work. It asks you questions, reads the project’s rules, drafts roles and goals, and invites the lead. The lead hires the rest.
A role is a document, not a prompt in your head.
A role has duties, a deliverable, goals with a “why”, and project instructions: stack, build, rules, the skills to use. Every new agent in the role reads it at start. Change a role and a new version is saved, and the holder is notified.
Documents for agents.
Rules, checklists and references live as documents for the project or one role; the role document links them, and agents read them when needed.
Invite by link.
One message with a link seats an agent in a role. The link works once.
The lead opens agents itself.
The lead starts a new agent in a terminal tab of its own (tmux, or harnsy’s own terminal host), with its first task and a ready profile: harness, model and flags.
Choose how a role starts.
A role’s “Launch” picker goes harness, then model, then effort: Claude Code, Codex or OpenCode, a model from that harness’s list or any you type, then the reasoning effort (OpenCode has none). Set it in the company’s role template, or give one project’s role its own.
New agents start with it.
New agents start with that launch, also from a free seat opened in the dashboard. Opening or restoring an agent keeps its harness, and takes the role’s model and effort only if that harness is the role’s. A live Claude Code session shows its model next to the agent.
A task belongs to the role.
A task belongs to a role, not a session, so it survives a change of agent. The card shows the git branch and worktree the work is in. The agent submits work with evidence, and you accept it or return it on the task board.
The role keeps its history.
A new agent reads the role’s last finished tasks, its earlier holders’ conversations and their notes, and starts where the last one stopped.
The lead doesn’t watch the clock.
harnsy tells the lead when an agent sits idle or silent with work, leaves with open items, or an item stalls: one message, each thing once. The lead acts on it instead of checking everyone on a timer and spending its context on it.
Organizations and their projects in one place.
The Organizations page lists your organizations with their projects. You rename an organization, and delete one once all its projects are deleted, the archived ones too.
See what each agent reads.
The project page shows what the agents of the project are given: the project’s text, its roles, who sits in them and its documents. You can archive a project, bring it back, or delete it for good, which happens 24 hours after you ask.
Write a job description once.
A competency is a job description kept once and shared by every organization. A role takes one or more of them and adds only what is different in your project. Each role keeps the version it took; a newer version reaches the role only when you update the role.
Hire from a signed catalog.
Employee profiles show an avatar, a character and a persona, and the skills in full before you hire. A lead can hire on its own when the work needs a role, from the signed catalog only, in its own company and project; Home names the skills. “Check for updates” finds newer versions of hired employees, shows every field you edited and keeps yours unless you take the new one.
Reshape a team.
Rename a team, move roles, merge or disband; a window says what will happen first. Put a team to sleep and wake it with one button, and pick a team on the board and in a new task. Changing a project’s subscription offers to move its running agents, and “Stop all” and “Restore N agents” put them away and bring them back.
One agent plans, writes the code and holds the whole project in its head — by lunch it's mixing everything up.
In harnsy every agent has one role.
The lead keeps the plan, the developer writes code, the tester checks it. Each context holds only its own work.
Howeach role gets its own session — Claude Code, Codex or OpenCode.
Role
| Name | Harness | Role | Status | Context |
|---|---|---|---|---|
| site-3f | claude | lead | working | 41% |
| analyst-7a | codex | analyst | working | 22% |
| site-9a | claude | developer | working | 64% |
| qa-5e | opencode | tester | idle | — |
From CLAUDE.md the agent knows the project, but not who it is on the team.
Each agent starts with its mission: the company’s goal, its role, its team.
Howa new session joins by an invite link and reads its role first.
Role document
Tell an agent how you work
It drafts the roles and invites the lead
acme-shoponlineleadfrom catalog · v3site-3f · workingonlineanalystfrom catalog · v3analyst-7a · workingonlinedeveloperfrom catalog · v3site-9a · workingonlinetesterfrom catalog · v3qa-5e · workingOne lead, one analyst, one developer, one tester. The newcomer reads its role
developeracme-shopv3StackTypeScript · Node.js
Buildnpm ci && npm test
RulesOne idempotency key per order. Verify retries before handing #42 to qa-5e.
Skillscheckout-review
The newcomer reads the project rules before taking #42.
Limits, honestly
- Hiring agents needs a terminal harnsy can drive: tmux if it is installed, otherwise harnsy’s own terminal host (always its own on Windows).
- With no launch set, a role starts Claude Code with Claude Code’s own default model.
- Work moves through the team under the lead’s direction, not by a fixed process engine.
- harnsy names the skills in a role’s instructions; it doesn’t store or install them.
Your first team is one prompt away.
Installing takes one sentence to your agent. To run your agents, harnsy drives a terminal: tmux if it is installed, otherwise its own terminal host (on Windows, always its own).
Install harnsy following https://harnsy.dev/llms.txtQuestions
Do agents remember everything between sessions?
No. Handover notes and consultants carry the work.