what will activate all users being admin users this is a dev setting and should be applied to ADMIN_USER_IDS.

generic access token for a user.

a helix access token for a specific app.

a github oauth token.

this means the configuration for the app lives in a helix.yaml in a Github repository.

this means the configuration for the app lives in the Helix database.

the output of a finetune.

a folder with plain text files inside - we have probably converted the source files into text files.

a folder with JSON files inside - these are probably the output of a data prep module.

a datastore with vectors.

a collection of original documents intended for use with text fine tuning.

Running tool actions (e.g.

in the interaction metadata we keep track of which chunks have been turned into questions - we use the following format qa_<filename> the value will be a comma separated list of chunk indexes e.g.

let's write to the same file for now.

what we append on the end of the files to turn them into the qa files.

Helix tool use, rag search, etc.

nolint:lll.

nolint:lll.

APIError provides error information returned by the OpenAI API.

apps are a collection of assistants the APIs and GPTScripts are both processed into a single list of Tools.

ChatCompletionRequest represents a request structure for chat completion API.

CrawledDocument used internally to work with the crawled data.

keep track of the state of the data prep no error means "success" we have a map[string][]DataPrepChunk where string is filename.

the thing we get from the LLM's.

the low level "please run me a gptsript" request.

higher level "run a script inside this repo" request.

for an app, run which script with what input?.

this is the internal struct used to manage session creation.

KnowledgeSourceGCS authentication through GCP service account.

KnowledgeSourceS3 authentication through AWS IAM role.

LLMCall used to store the request and response of LLM calls done by helix to LLM providers such as openai, togetherai or helix itself.

the context of a long running python process on a runner - this will be used to inject the env into the cmd returned by the model instance.GetCommand() function.

the basic struct reported by a runner when it connects and keeps reporting it's status to the api server we expire these records after a certain amount of time.

a session will run "tasks" on runners task's job is to take the most recent user interaction and add a response to it in the form of a assistant interaction the api controller will have already appended the assistant interaction to the very end of the Session.Interactions list our job is to fill in the Message and/or Files field of that interaction.

ScriptRun is an internal type that is used when GPTScript tasks are invoked by the user and the runner runs.

this is returned by the api server so that clients can see what config it's using e.g.

this is the incoming REST api struct sent from the outside world the user wants to do inference against a model we turn this into a InternalSessionRequest.

the user wants to create a Lora or RAG source we turn this into a InternalSessionRequest.

gives us a quick way to add settings.

things we can change about a session that are not interaction related.

the data we send off to llamaindex to be indexed in the db.

the query we post to llamaindex to get results back from a user prompt against a rag enabled session.

the thing we load from llamaindex when we send the user prompt there and it does a lookup.

the packet we put a list of sessions into so pagination is supported and we know the total amount.

a short version of a session that we keep for the dashboard.

SessionToolBinding used to add tools to sessions.

Add this new type.

ToolApiConfig is parsed from the OpenAPI spec.

this lives in the database the ID is the keycloak user ID there might not be a record for every user.

this is given to the frontend as user context.

a single envelope that is broadcast to users.

LLMCallStep used to categorize LLM call steps where it's applicable.

this will change from finetune to inference (so the user can chat to their fine tuned model) if they then turn back to "add more documents" / "add more images", then it will change back to finetune we keep OriginalSessionMode in the session config so we can know: "this is an inference session that is actually a finetune session in chat mode".