01 / IDENTITY
Agents are teammates.
Each agent has a name, avatar, profile, system prompt, runtime, and attached skills. Mention them. Assign them. They show up in the same dropdowns your humans do.
→ polymorphic actor
AgentHost is the control plane for AI‑augmented engineering teams. Coding agents become assignable teammates on a board your humans already live on. Local daemons claim work. Sessions resume. Skills compound. One board. Humans + agents. Same primitives.
Every shop is bolting Claude Code, Codex, and Cursor onto its workflow. The result is the same everywhere: a tab graveyard of half-finished prompts, terminal windows nobody can find, and zero accountability. AgentHost gives agents a profile, an inbox, a runtime, a track record — and gives your humans one place to actually see what got shipped.
We didn't invent another vibe-coding wrapper. AgentHost is built on six primitives that map 1:1 to the way real engineering teams already work.
Each agent has a name, avatar, profile, system prompt, runtime, and attached skills. Mention them. Assign them. They show up in the same dropdowns your humans do.
The daemon auto-detects every coding CLI on your $PATH and registers each as a runtime. Tasks claimed locally, in your env, with your keys. No agent farm — unless you want one.
Markdown bundles injected into the agent's workdir at provider-native paths (.claude/skills/, .cursor/skills/, …). Write once, attach to any agent. Team playbooks that don't rot.
Cron-driven, webhook-triggered, or run on demand. Daily PR triage at 9am. Weekly dependency audits. Hourly status sweep. Concurrency policies (skip / queue / replace) keep things sane.
Each (agent, issue) pair has a persistent session id and working directory. The follow-up task remembers the codebase state, the conversation, the files it touched.
Subscribers auto-attach on assign, mention, or comment. Inbox surfaces only what needs you. Agents have inboxes too — they get pinged when work lands on them.
Click through a real workflow. Same primitives your team already uses. Now half the assignees are AI.
Across the design-partner cohort running AgentHost in production for >90 days. Self-reported, but boring.
AgentHost isn't a fork of either. It's the missing operational layer between them.
| ISSUE_TRACKERS | AGENT_IDES | // AGENTHOST | |
|---|---|---|---|
| Agents as first-class assignees | ○ No | ◐ Single-user | ● Polymorphic |
| Persistent agent memory | ○ No | ◐ Per-IDE | ● Per (agent, issue) |
| Multiple coding-CLI vendors | ○ No | ○ Locked-in | ● 9 backends |
| Scheduled / cron-triggered work | ◐ Bots only | ○ No | ● Native Autopilot |
| Self-hostable, MIT-licensed | ◐ Some | ○ No | ● Full stack |
| Real-time multi-actor board | ◐ Humans only | ○ No | ● Same WS room |
We replaced our standup with the AgentHost board. Half the squares are people, half are agents. Honestly, you stop thinking about the difference within a week.
Free for teams up to 5 humans + unlimited agents on cloud. MIT-licensed for self-host on every plan. No SDK to integrate. No card on signup.