Avery
RuntimeUse casesPricingHelpBlog
← All postsBlog

Why we built 63 connectors (and which ones we'd build next)

2026-06-18 · Avery NXR

Avery NXR ships with 63 connectors out of the box: 15 OAuth + 48 API-key, across 13 categories.

That's a specific number. Not 50, not 100. Behind it is a real prioritization story about which integrations matter most for operational AI agents — and which ones we deliberately skipped.

Here's the connector economics behind Avery NXR.

What "63 connectors" actually includes

The 15 OAuth connectors handle the common B2B SaaS surfaces that need real authentication:

Google Workspace, Microsoft 365, Slack, GitHub, GitLab, Linear, Notion, HubSpot, Jira, Asana, Zoom, Calendly, Discord, Loom, Stripe.

The 48 API-key connectors handle services that authenticate via API key — which covers most everything else operational agents need:

Email providers (Resend, SendGrid, Postmark, Mailgun), SMS (Twilio, Sinch), search (SerpAPI, Tavily, Brave), market data (Alpha Vantage, Finnhub, Polygon), commerce (Shopify, BigCommerce, WooCommerce), vector DBs (Pinecone, Weaviate, Qdrant, Chroma), traditional DBs (Postgres, MySQL, MongoDB, Redis), AI providers (OpenAI, Anthropic, Google, Mistral via BYOK), monitoring (Datadog, New Relic, Sentry), and ~25 more.

How we picked the 63

Three factors:

Factor 1: How often does the connector show up in agent workflows? We built a list of the operational agent workflows our target audience runs. Then we counted how often each external service appeared as a step.

Top of the frequency list: email send, calendar read, Slack post, CRM read/write, file read, calendar event create. We made sure connectors for the top 30 were rock-solid before adding any more.

Factor 2: Is the auth model standardized enough to be reliable? OAuth connectors are harder to build correctly than API-key ones. We capped OAuth at 15 because each one requires substantial maintenance work. We focused OAuth on services where:

→ The integration is critical (you genuinely can't substitute) → The OAuth flow is well-documented → The service has API stability guarantees

API-key connectors are simpler — bring an API key, we use it. So we could go broader on API-key (48) without proportional maintenance overhead.

Factor 3: Does the service have an alternative that's already covered? When we considered adding HelpScout, Front, Zendesk, Intercom, Freshdesk as separate connectors, we noticed: they all do roughly the same thing (customer support). Customers usually pick one. We added Zendesk + Intercom as representative coverage. Others can use generic HTTP/webhook capability.

Why we DIDN'T just build "1,000 connectors like Zapier"

Connector count is a vanity metric.

Zapier reports 8,000+ integrations. Most teams use 5-15 of them. The other 7,985 are noise that makes browsing harder.

We optimized for: when you sit down to build an agent, the connector you need is there + it works reliably. 63 well-built connectors beats 1,000 sketchy ones for that goal.

If a service we don't have is critical for a customer, the generic HTTP/webhook capability covers it — they configure an HTTP node with auth headers, done. Not as polished as a native connector but functional.

The "free vs. paid tier" connector question

Some agent platforms gate connectors behind tiers. Slack on the free plan, Salesforce on the paid plan.

Avery NXR Free Desktop includes all 63 connectors. No gating.

Why: because the value of agents comes from chaining multiple services together. Gating connectors behind tiers means users can't even SEE what the platform can do until they pay.

We'd rather users see the full platform on Free Desktop and decide whether they want Pro for the deployment + Consult Mode + collaboration features. Not lock them out of basic functionality.

What this looks like for a user

When you install Avery NXR, you see all 63 connectors. To use any one:

→ Click connector → authenticate (OAuth) or paste API key (API-key) → Connection is stored in OS keychain → Reusable across every agent

Once a service is connected, every agent can use it. The connection isn't per-agent.

This matters because connector setup is friction. The user does it once. After that, agents can use any combination of connected services.

Which connectors we'd add next

Based on customer feedback + agent workflow patterns, the top candidates for next batch:

→ Salesforce. We have HubSpot. Salesforce is the gap for enterprise customers. Complex auth model, but worth the engineering. → Microsoft Teams. We have Slack. Teams matters for Microsoft-shop customers. → Trello + ClickUp. We have Linear / Asana / Jira. Trello and ClickUp round out PM coverage. → Google Sheets. Cousin to Excel sync. Many customers ask for this specifically. → Airtable. Half-database half-spreadsheet. Heavy use in small biz tooling. → Webflow + WordPress. Content surfaces matter for marketing agents. → Calendar providers beyond Google. Outlook calendar, Cal.com, others.

We could ship 5-10 more connectors per quarter without diluting quality. Each requires testing against real workflows, documenting edge cases, monitoring deprecations.

The "but you don't have X" objection

Every prospect we talk to has at least one service they need that's not in the 63. That's fine.

The path: use the generic HTTP/webhook capability. Configure auth headers. Test against their API. Done in ~20-30 min for most services.

Not as smooth as a native connector. Works.

When a service comes up in customer requests repeatedly, we promote it to native. The customer requests feed our roadmap directly.

What connector economics tells you about a product

When you evaluate an AI agent platform, look at the connector list. The number, the curation, the gaps.

Too few connectors = platform is early or focused too narrowly. Too many connectors = platform optimized for marketing claims, probably has reliability problems. Curated 50-100 = platform made deliberate choices about depth vs breadth.

Avery NXR's 63 is in the third camp. We picked carefully, built each one well, ship them all on every plan.

The connectors are working for ~95% of customer use cases. The other 5% use HTTP/webhook. The math has held up.

→ avery.software — Free Desktop tier. All 63 connectors. No gating.