Shaped review

Shaped is a managed platform for search, recommendations, and retrieval for AI agents. It combines data ingestion, embeddings, ranking, and feedback-driven learning in one queryable system, rather than positioning itself as just a search API or an embedding model.

Its hybrid search combines semantic and keyword retrieval. Ranking can use user and item context, behavior, and business rules; interactions can then feed back into future results. For agents and assistants, Shaped offers contextual retrieval and memory. ShapedQL provides a SQL interface to retrieve, filter, score, and reorder results.

The platform also ingests batch and streaming data through native connectors and unified schemas. The published product information doesn’t name those connectors, so teams should verify that their sources are supported before treating this as a replacement for an existing ingestion stack. The buying question is whether managed relevance is worth the monthly floor and usage charges—not whether Shaped has the basic parts of a search system.

The clearest advantage is scope. Shaped brings ingestion, retrieval, ranking, and feedback learning into one managed platform. For a team currently stitching together separate services for search, recommendations, and agent retrieval, that could reduce the number of systems it needs to operate. Whether it actually replaces those tools depends on connector coverage and the team’s ranking requirements; the published information doesn’t list supported sources or implementation details.

The feature set also covers more than text relevance. Shaped can rank using user and item context, behavioral signals, and business rules, while its hybrid search combines semantic and keyword results. That gives teams a stated path to use both query meaning and exact-term matching, then refine results with behavior.

ShapedQL is another practical point: a SQL interface for retrieving, filtering, scoring, and reordering results may be easier to work into teams’ existing query habits than a proprietary workflow. Agent retrieval and memory sit alongside standard search and recommendations, which may matter if those workloads share data. The pricing page lists a free Starter plan with console access and community support, giving teams a way to explore before committing to the Standard minimum.

The main obstacle is price clarity. The pricing notes list Standard at a $500 monthly minimum, with $42 per million queries, $5 per hour for GPU training or encoding, and $0.75–$2.25 per GB for storage. Annual pricing is not published. Pro support is listed separately at $500 per month, while Enterprise pricing is custom. The materials don’t explain how usage is counted against the minimum or provide a worked bill, so estimate your workload before comparing costs with a point solution.

Starter’s allowance is also unclear: the pricing lists $100 in free credits and, in the same line, “also listed as $300/month.” Those figures don’t reconcile in the published details. A 14-day free trial is listed, but its terms and how it relates to the free credits are not specified.

The published information also gives no published query, storage, or ingestion caps for Starter, and no connector list or performance figures. Enterprise offers a 99.95% uptime SLA, dedicated engineering, and 24/7 support as available options, but pricing is custom. Teams that need predictable costs or proof of fit should resolve those gaps before switching.

Shaped lists a free Starter plan and a 14-day free trial. Starter includes console access, community support and free credits, but the pricing notes conflict: they describe the credits as “$100 in free credits” and also list “$300/month.” Confirm the credit amount and any limits before treating Starter as a usable budget.

Standard is the big step up: a $500 monthly minimum, then $42 per million queries, $5 per hour for GPU training or encoding, and $0.75–$2.25 per GB of storage. Those metered charges can push the bill above the floor. Pro support adds $500 per month. Annual pricing is not published.

Enterprise pricing is custom; the pricing notes say a 99.95% uptime SLA, dedicated engineering and 24/7 support are available on custom terms. The headline $500 minimum is therefore a poor fit for teams that need a predictable, low monthly cost. Get a written estimate based on query volume, training, storage and support before switching.

A move to Shaped means lining up your batch or streaming data with its unified schemas, then deciding how to represent user and item context, behavioral signals and business rules for ranking. Its product description lists native connectors, ShapedQL, hybrid search and feedback learning, but the materials here do not specify migration tooling, supported export formats or how much implementation work a move takes. Ask for those details before committing.

The same gaps matter when leaving. We found no published documentation on exporting data, embeddings, ranking configuration or feedback history, so do not assume those transfer cleanly. Check which records and settings you can retrieve, in what format, and whether your existing search or recommendation stack can use them.

We also have no Reddit posts or comments in the published material to support claims about why customers leave. The clearest buying risks here are the $500 Standard minimum, usage-based charges and unpublished annual pricing—not verified user complaints.