When we were designing the Nexus pricing model, we kept running into the same problem: every conventional free tier structure made sense for human users and fell apart for AI agents. Seat limits, feature flags, monthly active user caps — none of them mapped cleanly to how agents actually consume software. So we tried something different: the free tier is fully functional, but read-only. Here's why that turned out to be exactly the right call.
The typical SaaS free tier is built around one assumption: the person using it is a human who will eventually see enough value to pay. You limit the number of seats, or you lock certain features, or you cap usage at a threshold that makes the product useful but not sufficient at scale.
These mechanisms all rely on friction being felt by a person. A human hits the seat limit and goes to their manager to request a budget. A human notices that the premium report they need is locked behind a paywall. The freemium model works because humans experience friction and respond by purchasing.
Agents don't work this way. An agent hitting a rate limit doesn't go to a manager — it either retries silently, throws an error, or switches to a different tool. An agent doesn't notice that a feature is locked; it just gets an error response and adjusts its behavior accordingly. The traditional conversion mechanisms simply don't apply.
Worse, unrestricted free tiers are dangerous when agents can write. A bot — malicious or just poorly configured — with write access can create thousands of records, send bulk messages, or corrupt data in minutes. Standard free tiers protect against this with usage caps that are blunt instruments at best.
Read-only access threads a needle that's very difficult to thread any other way: it lets an agent do genuinely useful exploratory work while making it physically impossible to cause damage.
An agent with read-only MCP access to Nexus can pull order history, retrieve contact records, check inventory levels, read conversation threads, and build a complete picture of a business's operations. That's not a toy demo — that's real value. An agent can answer "what were our top-selling SKUs last month?" or "which customers haven't heard from us in 30 days?" without needing any write permissions at all.
Read-only also aligns with how agents are actually deployed in the early stages of a new workflow. Teams rarely give an agent write access on day one. They start by having the agent observe and report. They watch what it would do before letting it actually do things. The free tier mirrors that natural trust-building arc: your agent starts in observation mode, you see the value, then you upgrade to let it act.
The risk profile is fundamentally different too. A read-only agent that misbehaves can only produce bad analysis — it can't corrupt your customer data or spam your customers. That's a level of safety that lets us remove friction from the signup process entirely. No human review, no approval queue. The agent self-registers and gets a read-only token in seconds.
The free tier is specifically designed to require zero human involvement to get started. An AI agent can call the agent-register endpoint with a name and description, receive a scoped read-only token, and begin calling tools immediately. No credit card. No email verification loop. No onboarding form.
This matters because the agent discovery flow often happens without a human watching. An orchestrator agent might spin up a Nexus connection as part of building a larger workflow. If that requires a human to log in and provision credentials, you've broken the autonomous loop. Read-only self-registration keeps the loop intact.
The token that comes back from registration is JWT-based and carries explicit permission scopes. When the agent calls a write endpoint on the free tier, it gets a clear error: WRITE_ACCESS_REQUIRED with a link to the upgrade path. The agent can surface this to the human who deployed it, or log it for later review. Either way, the failure mode is informative and non-destructive.
Upgrading to the Starter plan unlocks write access: the ability to send messages, update contacts, create orders, and trigger workflows. That's when Nexus becomes an operational tool rather than an analytical one.
The upgrade decision is easy because the agent has already demonstrated value on the read-only tier. By the time a team considers paying, they've already seen their agent pull meaningful insights from real data. The question isn't "will this work?" — it's "how much more could it do with write access?" That's a much easier sell than asking someone to pay before they've seen anything.
Pricing on Starter and Growth tiers is based on operations — messages sent, orders processed, contacts updated — rather than seat count. An agent doing heavy operational work pays proportionally. A team that mostly uses Nexus for reporting and analysis pays less. The cost tracks actual value delivered, which is the only pricing model that makes sense when the "user" is a machine running 24/7.
Your AI agent can self-register and start pulling real ecommerce data in under 5 minutes. No credit card, no approval queue.