How Sirro searches saved work
A look at the search behind a library of named assets for coding agents.
A coding agent often searches Sirro with a name already in mind. Someone asks it to use the navbar they saved last month, retrieve a prompt that worked well, or find a pattern from an earlier project. Each of those pieces was deliberately saved and named. Sirro's search starts from that small but valuable clue.
Sirro stores assets in PostgreSQL. When an agent lists a workspace, the service reads asset metadata, selects the latest version of each named asset, and filters results against the search query. The current match is a case-insensitive text match across the asset's name, type, category, project, description, and tags. The full saved content is retrieved after the agent chooses an asset; it is not searched by the list operation. Giving a saved piece a clear name and description helps the next agent find it quickly.
Search results stay compact so an agent can decide what to read. It can inspect the matching names and descriptions, then fetch the full asset that belongs in the current task. This matters because every tool result takes space in the agent's working context. Loading a whole library at once would make a simple lookup harder. The two-step flow lets the agent discover what exists before bringing the useful content into view.
PostgreSQL remains the source of truth for the saved assets and their versions. The list search reads those records directly, within the authorized workspace. There is no separate vector store or search index to update when an asset changes. A saved version remains available for a precise lookup, while the list presents the latest version under each name.
This is the search Sirro uses today. It does not currently use PostgreSQL full-text ranking, trigram indexes, or embeddings. As beta libraries grow, we will look at query patterns, missed results, and search latency to decide where better ranking or broader matching would help. For now, a good name, a useful description, and a focused retrieval flow give coding agents a direct way back to the work people chose to keep.
If you want to see how it works with your own projects, join the closed beta at beta.sirro.dev, save a piece of work under a name you will remember, and ask a connected agent to find it later.