An Airbyte alternative for AI agents: managed data and SQL

Compare the product your agent actually needs

Airbyte has two relevant offerings: Data Replication moves data into a destination you choose, while Airbyte Agents provides agent connectors and a managed Context Store. If you are building a business-data agent, compare the complete path from source authorization to a useful answer.

Airbyte’s Context Store supports filtering, sorting, and aggregation over a curated selection of source entities and fields. Some connectors also support semantic search. Its agents can use direct API requests for current records, unsupported entities, and writes. Those capabilities matter when comparing it with a read-only data layer.

Combined’s appeal is the SQL workflow: discover the granted logical datasets, inspect their schemas, check freshness, and query them with the agent you already use. Your team can focus on business questions without separately operating the ingestion destination and its agent-facing query service.

Start with the business question, then compare the answer

A useful evaluation question crosses the boundary between two applications. For example, a CRM can tell you which companies have open opportunities, while billing records can show their paid invoices. Giving an agent access to both is only useful if it can connect the records correctly.

Using the granted CRM and billing datasets, find companies with open
opportunities and summarize their paid invoices from the last 90 days.

First discover the schemas and check source freshness. Establish a
shared company identifier before joining; don't join on company name
alone. Aggregate each side before joining so multiple deals don't
duplicate invoice totals. Keep currencies separate.

Return at most 20 companies, with open opportunity value, paid invoice
total, the time window, source timestamps, and the query receipt.
If a matching key or required field is missing, explain what is needed.

In Combined, the agent uses catalogue discovery and describe_dataset to learn the available fields, then query_sql for the bounded analysis. The important output is a result you can inspect: a clear join, consistent time bounds, separate currencies, and an explanation of the data’s age. This is a worked evaluation prompt; adapt it to the datasets you have connected.

Context Store or SQL for account totals?

Airbyte's Context Store supports native aggregation, and its cross-system examples show an agent combining results from different connectors. Evaluate that path alongside SQL using the same account mapping, qualifying records and expected totals. The deciding question is whether the complete calculation is inspectable and repeatable.

Use the Salesforce–Stripe workflow, its runnable SQL fixture and expected output as the answer contract. Aggregate opportunities and invoices independently, then join by account and currency. This synthetic fixture contains only USD.

The paid column sums amount_paid only for invoices with status = paid and a paid timestamp from June 12 inclusive to September 10 exclusive, 2026 UTC. Atlas's $500 partial payment on a still-open invoice is excluded; the answer is $3,000, not $3,500. The fixture converts USD minor units by dividing by 100. This measures amounts paid on qualifying paid invoices, rather than every cash receipt in the interval or net revenue. Stripe documents invoice status, payment amounts and remaining balance separately.

Keep open pipeline and outstanding balances as current-state measures. At the fixture's September 10 snapshot, $500 of Atlas's $700 remaining balance is overdue and $200 has no due date. The void invoice and the paid invoice outside the interval do not qualify.

1. Identity and grain. Map each billing-account/customer pair to the Salesforce account, allowing multiple customers per account. Aggregate each side before joining. Preserve accounts with open pipeline; a missing mapping or incomplete invoice history means unknown, not zero.
2. Coverage. Verify required objects, fields, source-account scope and the full reporting interval. Airbyte exposes per-entity status, counts and update times; Preview can contain partial data. Confirm any fetched pages are complete and that a native aggregate covers the intended records. See the status definitions.
3. Evidence. Keep the filters, intermediate aggregates, mapping and final calculation. In Combined, use discovered granted relations and retain the SQL receipt. A committed sync establishes when data landed; separately verify interval coverage and whether later source changes are included.

Choose Combined when you want this calculation expressed as SQL over managed, synced business datasets. Open the Salesforce–Stripe plan and download the platform evaluation worksheet to carry the same requirements into your trial. The synthetic result is an evaluation contract, not an observed comparison of either vendor's production performance.

Start smaller if one source is enough. The HubSpot tutorial walks through pipeline analysis; the Claude Code guide covers the MCP connection. Use the same question and permitted fields when evaluating another platform.

Compare changed records with agent operations

Combined’s public rate is $5 per million monthly active records after the trial, with no public base charge or minimum. MAR measures changed records at the committed data boundary. The initial cardless allowance is 5 million MAR. Query budgets are separate: the default monthly input budget is 100 GiB per account, with a 256 MiB estimated-input ceiling per query.

Airbyte Agents meters agent operations, or AOs, which represent processing work rather than a fixed number of requests. Bring your own agent through MCP, API, SDK, or CLI and Airbyte counts tool-call work, not that external agent’s reasoning. Airbyte-hosted Chat includes reasoning; local open-source SDK connector calls do not consume AOs.

Airbyte’s billing documentation lists Free and Individual as self-serve, with Team and Custom sales-assisted. Free refreshes the Context Store hourly for two weeks, then daily; Individual and Team refresh hourly.

One million MAR and one thousand AOs describe different work. Estimate your source changes, query sizes, and agent activity before comparing bills. Combined is straightforward to size around changing business records; Airbyte Agents requires sizing the agent’s operations. A trial allowance and a recurring free plan also serve different stages of the evaluation.

Try Combined with one source and your existing agent

You can evaluate Combined alongside your current stack. Pick one question worth answering, confirm the source’s Ready status and required object coverage, connect it, and sync the selected datasets. Grant the agent access to that scope and use the account-specific connection instructions in the Access workspace.

Choose Combined when you want managed business data with explicit SQL over granted relations. If the same agent also updates CRM records or sends messages, pair that read path with the appropriate action integration. Keep each source’s freshness and the permissions required for the job visible in the workflow.

See how Combined brings business context to your agent, then start with the cardless allowance below. Get one useful answer, inspect the result, and expand to the next source when the next question needs it.

Sources and further reading

Explore the documentation behind this guide. Product details checked on September 11, 2026.

• Combined pricing
• Combined MAR and query budgets
• Combined MCP tools and query workflow
• Airbyte Context Store
• Airbyte cross-system examples
• Stripe Invoice object
• Combined ingestion and provenance
• Airbyte agent operations
• Airbyte Agents billing and plans
• Airbyte Data Replication plans