The user can interact with CogniX, start a new chat in two different ways (in the future will be expanded)

1. Chat with your documents (requires_llm = true)
2. Search only (requires requires_llm=false)

The user asks a question (query): "How do I do get attention in Collaboard"

Server receives the chat

following are business logic methods:

method name strem_response
argument query (string)
returns the stream

strem_response calls vector_search
method name vector_search
argument query (string)
returns a list of documents

It performs a query against Milvus, with userid and tennantid
The query result will contian all the matching documents inside Milvus.

if equires_llm == true
It retrives the information regarding Persona, prompt and llm

Top X documents retrieved from Milvus query will be sent to the LLM (LLM Endpoint + eventually api_key) embedded into the prompt

LLM will stream the answer which will be streamed back to the client

if equires_llm == false
will send back only the list of documents

Chat and response is saved into chat history

https://milvus.io/docs/configure_nats.md

now the code is duplicated between connector and orchestrator

you need to create a package so the same code can be accessed by both

you can even add to an existing package

Hi Noel,
For development purposes only (this will not be allowed in prod) we need to access Milvus as we do in our Htezner server.
We need, from our local PC to be able to connect to milvus and perform Insert, select etc.

can you help, please?

We need to define

• the message shall be taken by only one of the listener
• number of retries before sent to dead letter
• dead letter persistence (set max time?)
• persistency with jetstream
• serialization protobuf if possible

In our containers
We will have different config map file, and also some globals.

From the back office we will have the list of all config maps (K8 API) and then we will be able to select which one to edit.
THe UI will be able to save back config map changes

Once the changes are applied, there will be the config map reloader operator that will rolling upgrades all the pods that have the annotation to listen to the conmfigmap file change

Talk with @noelhermans about this, pls

filled with data above, clicked add several times nothing happens

look at the screen bottom left
i cannot login, looks like an issue with token

finalize all chat apis as discussed

if the value in the json is not valid like this this case

{"search_for_sitemap":false,"site_map":"","url":""}

API shall send and error message invalid json or something else

@noelhermans can we have grafana inside the cluster?

I need for a demo...

Change to "+ Connector"

image

Remove connector credentials form UI, API and database

Create two classes that inherit from the same interface
create a config map param:
Key: CHUNKING_STRATEGY
Values, STATIC, LLM

Implement first static:
It chunks the text with a fixed length (param coming from config map)

Key: CHUNKING_STATIC_CHARS
Values, (int)

Key: CHUNKING_STATIC_CHARS_OVERLAP
Values, (int)

LLM:
It took a big chunk (compatible with the LLM chat length)
Define in the LLM table the max text size

then make a query to the llm with a prompt (ask Gian for the right prompt)

We need to be able to set the log level, and any other open telemetry parameter in the config map

User should registered using endpoint https://github.com/gen-mind/ibv-docs/blob/main/IBV.Auth/Authenticating-using-local-accounts.md#user-registration

After complete registration UI should send a request api/tenant/registration . user and tenant data will send in Authentication token that receive from Auth service.
Backend should check if the tenant already exists.
Backend should prepare database tables for work.
add tenant
add user
add predefined persona for tenant
add embedded models for tenant
configure api key ( at least add openAI api key )

Question:
Need to clarify use case when tenant already registered
will the frontend communicate with Auth service directly ?
whether the admin should invite new users.
or backend must configure the user at first login

if i make a fresh install
i open the app, i never logged in and I see the following errors

We need to remove from the cluster all the old components (vespa, postgress, any config map related to it, and python containers)
We need the new web app to be shown at rag.cognix.com

Sync with Vadym and @apaladiychuk pls

rename to refresh frequency (seconds)

please implement the design of the login page, refer to the approved figma design

we need to trace the amount of gb/tb the user uses

relational database + vector database database + storage

we need to store in the user profile and in the tenant profile max storage allowed

after that they start paying an amount per gb

try now to add connectors, close the poup and open again

the list is always in a different order,it shall be always in the same order

Add connector - connector list shall be always the same

1. create dockerfile for api
2. create docker compose for: db migrator, api, postgres db

Setup authentication for

1. Milvus
2. Cockroach
3. Pulsar
4. etc.

Once done inform @apaladiychuk

Thank you.

Introduce Slimtoolkit in our pipeline for Go lang and python

There is not displayed response on send message request .
Backend send several events
message
error
dcuments .

documents - still fake data
error - if has error in request data from milvus
message - response from AI

Operator for restart pods with label when config map is changed

after clicking to the pause button

have a look a the competitor connector to have an idea

write design specs for connectors.
my requirements.

• all config params read from config map
• use open telemetry for distributed tracing (need to pass the context to nats) and metrics
• envet driven design
• base class connector from which every connector inherits
• there are personal connectors and tenant connectors

after the design is ready work on #3

if a user wrowse
rag.cognix.ch
This is the first error it sees, as soon as arrived :) :) :)

The llist need to expose the connector name otherwise will be impossible to identify the right connector!!!

please have a look at the video attached

https://drive.google.com/file/d/1IHQh4LDmTau24dnV-Lf33JEkQkWH06MN/view?usp=sharing

see attached video, pls

https://drive.google.com/file/d/1OXDAI3GEcrNjP8CKCgIxK41nivtp-iRr/view?usp=drive_link