Roles — AgentNavaKit

Pick by use case

A bot on your website or app, serving many users at once

You're embedding an agent on a public site, in a chat widget, or behind an API that many end users will hit. Each user has their own conversation; the agent doesn't carry state across users.

Pick role: 'freelancer'. Fan-out friendly, fast cold-start, scales horizontally. The right shape for support bots, sales assistants, FAQ agents, scheduled report runners, anything that talks to many people in parallel.

A specialist that does a role of work for you

You want an agent that owns a job — a code reviewer, a researcher, a finance analyst, a content drafter. It builds up knowledge as it works, has its own tools and workspace, and you come back to it over time the way you'd come back to a contractor.

Pick role: 'teammate'. Persistent workspace, real filesystem and shell, remembers what it's done from one conversation to the next. Hire more than one and they work as a team — sharing one workspace, with a Manager coordinating. See Teams & workflow for the day-to-day.

There's a third role — manager — but you don't pick it directly. The platform auto-spawns a Manager when a workspace has two or more Teammates, to coordinate the team. See Managers below.

Freelancers — hosted bots

An agent that serves a stream of unrelated end users. Each conversation is its own thread; the agent has no shared memory across users.

Scales horizontally. 1 user or 10,000 users hitting it in parallel — same shape, same cost-per-turn.
What it has: declarative HTTP tools, MCP servers, web search, your knowledge base, widgets, subagent delegation via the Task tool. Everything a chat-shaped agent needs.
What it doesn't have: no Bash or filesystem. Nothing is persisted between users. The agent itself doesn't accumulate state — what each user discusses stays in their session and is invisible to everyone else.
Per-user continuity is built in. When a user comes back next week, they pick up the same thread (history, prior context). Across users, full isolation.
Cost shape: cheapest of the three. Fast cold-start; you pay roughly per-turn.

Examples: a customer-support bot on your marketing site; a scheduled "draft my weekly report" agent that fans out across thousands of users; a public-facing FAQ assistant; a widget you sell to other websites.

Lifecycle: the configure → deploy-to-test → deploy-to-prod flow described in Operate → Deploy and the Deploy & Embed guide is the Freelancer lifecycle. You push a spec, deploy a version to a hosted URL, then embed that URL on your site. Versions, rollbacks, custom domains, and the embed snippet are all Freelancer concepts.

Teammates — specialists in a team

Agents that do a job for you. They live in your workspace, share one persistent filesystem, and a Manager coordinates work between them. Think of it as hiring a small office of specialists — not deploying isolated chatbots.

The shape

One workspace for the whole team. Every Teammate (and the Manager) in your workspace lives in one always-on Linux environment with a shared filesystem. When one Teammate clones a repo, another can read it. When the researcher leaves notes, the analyst picks them up.
One folder per Teammate. Each agent you create has its own folder, its own conversation, its own tools, its own running processes. New ones appear automatically as you create them in the console.
Manager coordinates. Your first point of contact. The Manager answers what it can directly, hands the rest to the right Teammate, then synthesizes their replies into one coherent answer.
Real filesystem and shell. Everything a Freelancer has, plus Read, Write, Edit, and Bash. Your Teammates can apt install tools, clone repos, run databases, keep long-running processes, build up a real working environment over weeks.
Persistent across conversations. Files written today are there tomorrow. The relationship is continuous, like hiring contractors who remember everything they've done for you. (Teammates have one ongoing conversation per workspace member — not the multi-session fan-out model that Freelancers use.)
Delegation is local and instant. When the Manager hands work to a Teammate, it's an in-process handoff inside your workspace. No network hop, no inter-agent latency.

Examples: a code-review specialist that knows your codebase conventions, working alongside a docs-writer that reads the same checkout; a research analyst that builds up a library of source materials, with an analyst Teammate that runs models against them; a financial-modeling Teammate that maintains a working spreadsheet that other Teammates can reference.

Lifecycle: you don't deploy a Teammate to a URL — Teammates aren't hosted bots. You create one in the console (or push its spec via the SDK) and it joins your workspace from then on. You talk to it through the console, or through the workspace API. There's no test URL, no prod URL, no embed snippet, no custom-domain step. Updating a Teammate's instructions or skills just re-pushes the spec — the Teammate keeps the same workspace and conversation continuity.

Managers

You can't pick role: 'manager' directly. The platform creates a Manager automatically as soon as your workspace has two or more Teammates — its job is to be the front door and route work between them.

Same capabilities as a Teammate (workspace, filesystem, shell, tools).
Knows which Teammates are available and what each one's role is.
Synthesizes Teammate replies into one coherent answer back to you.

Standard vs Premium

Teammates work the same way on Standard and Premium — the team-of-specialists model, the shared workspace, the Manager coordinating, the persistent filesystem. What changes between plans is power, not shape.

	Standard	Premium
Default model class	Standard inference tier	More powerful inference tier by default — faster and stronger on hard tasks
Overall quality	Standard	Better quality across the board — your Teammates think harder and produce better work
Pricing	Lower	Higher — see agentnava.com/pricing

You can move between Standard and Premium without restructuring your Teammates — the spec, the team layout, and the workspace are unchanged. See Operate → Plan changes for the mechanics.

When does Premium pay off?

Premium is a quality upgrade. Reach for it when:

Your work needs the stronger model by default. Code reasoning, multi-document research, complex planning — the kind of tasks where the marginal model quality is the bottleneck.
You can tell when answers are better. If you find yourself regularly rewriting or correcting your Teammates' output, the quality bump on Premium pays for itself in saved iterations.

If Standard's quality is comfortable for the work you're doing, stay there. Pricing detail at agentnava.com/pricing.