osFoundry / Platform
Retrieval stages
Chain retrieval stages into a custom pipeline per path scope. Override the defaults at any level.
What it is
Retrieval stages control how osFoundry finds the context an agent needs before it answers. Retrieval is the step that pulls relevant passages out of your connected knowledge and data so a model responds from your material instead of guessing. Rather than run one fixed lookup for everything, Studio lets you chain stages into an ordered pipeline and attach that pipeline to a specific path scope, so the way retrieval works for one body of knowledge can differ from the next.
Inside osFoundry, pipelines are organized around scope. A default pipeline applies broadly, and any narrower path can carry its own. Because settings cascade, you override the defaults at any level: leave most of your workspace on the shared configuration and change retrieval only where a particular corpus needs it, without disturbing everything else. Retrieval runs inside the desktop workspace under the same BYOK, local-first model as the rest of osFoundry, so it uses your own keys and can operate over data that stays on your device.
This matters because a single global retrieval setting is always a compromise. A dense reference manual, a folder of short notes, and a data-heavy project each reward different retrieval behavior, and forcing them through identical stages leaves grounding weaker than it should be. Scoping the pipeline to the path where a corpus lives, and only overriding defaults where the material actually calls for it, lets you tune the retrieval that feeds your agents precisely instead of tuning the whole workspace at once.
Key capabilities
- Chain retrieval stages into an ordered pipeline that runs before a model answers
- Attach a distinct pipeline to a specific path scope so each corpus retrieves its own way
- Override default retrieval settings at any level instead of changing the whole workspace
- Cascade a shared default across most paths and customize only where the material needs it
- Runs BYOK and local-first in the desktop workspace, over data that can stay on your device
Use cases
- Knowledge engineer: gives a dense technical manual its own retrieval pipeline while a folder of short notes keeps the workspace default
- Agent builder: tunes retrieval for one project path so an agent grounds on that corpus without altering how every other agent retrieves
- Privacy-conscious team: keeps a sensitive data set on-device and configures its retrieval stages separately from cloud-connected knowledge
Frequently asked questions
What is a retrieval stage in osFoundry?
A stage is one step in how the workspace finds context before a model answers. You chain stages in order to form a retrieval pipeline, and that pipeline decides which passages from your knowledge and data get pulled in as grounding.
Can different folders or paths use different retrieval settings?
Yes. A pipeline is attached to a path scope, so one body of knowledge can retrieve differently from another. A default applies broadly and any narrower path can carry its own pipeline instead.
Do I have to reconfigure retrieval for my whole workspace to change one area?
No. Settings cascade, so you override the defaults only at the level that needs it. Most paths stay on the shared configuration while you customize the stages for a single scope.