is there a way to get historical/analytical data from lk cli? Sessions in a given period or something similar? What about recordings?

Thanks for reaching out, Francisco. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

lk egress list is empty

Thanks — I’m following up about your note that

lk egress list

returns nothing. I can help figure out why and get you access to recordings or session metadata. A few quick diagnostic questions to narrow this down: 1. Have you started any egress/recording jobs for the project you’re querying? (For example, did you call

startEgress

from an agent or your app?) 2. Are you running

lk

against the correct project / environment and authenticated as the same account that started the egress? (Which

lk

context or project ID are you using?) 3. Do you see any relevant webhook events (like

egress_started

/

egress_ended

) in your webhook receiver or logs? 4. How is egress configured for recordings in your setup — are you using cloud storage (S3/GCS/Azure) or local/agent uploads? 5. If you started a recording recently, did it complete successfully or fail? Any errors in

lk agent logs

or your agent application logs? While you gather that, here are immediate checks and steps you can run now: • Confirm project/context: run the

lk

command with the same project/context you used to start egress. • Check agent logs:

lk agent logs

to look for egress start/stop messages or upload errors. • Verify webhooks: if you have webhooks enabled, check for `egress_started`/`egress_ended` events and inspect

manifestLocation

/ output locations. • Inspect storage: if egress is configured to upload to S3/GCS/Azure, check that bucket/container for recent files or manifests. • If you triggered egress from an app, review that app’s logs for the

startEgress

request and its response (it will include an egress ID). Tell me the answers to 1–4 (or paste any error snippets you see), and I’ll suggest the precise next steps — for example, commands to find egress IDs, how to retrieve files from storage, or how to re-run a failed upload. , with interactive elements

and without egress nothing of this is available for programmatic access even though its stored in lk servers (I can see it in my lk cloud dashboard)?

Thanks — I hear your frustration: you can see recordings and observability data in the LiveKit Cloud dashboard, but

lk egress list

is empty and you can’t retrieve that data programmatically. I’ll clarify what’s happening and what you can do next. Short answer • The dashboard’s session/observability view (audio playback, transcripts, traces, logs) is part of Agent Observability and is visible in the Cloud UI. • That dashboard data is not exposed programmatically via the

lk

CLI or a public API without using egress. • To get recordings or session artifacts programmatically, you must export them with Egress (which uploads to your configured storage) or capture the data yourself via webhooks/agent logging. Why you see it in the dashboard but not via

lk

• The Cloud dashboard stores observability data for playback and debugging, but there is no CLI/API that directly returns that stored observability/recording data unless you export it with Egress. •

lk egress list

only shows egress jobs. If no egress jobs were started, it will be empty even if the dashboard shows session data. What you can do now (options) • Use Egress to export recordings and artifacts: ◦ Start egress (e.g., room/track composite) in your app or agent so recordings are written to S3/GCS/Azure or another destination. ◦ Then

lk egress list

will show the egress sessions and you can download files from your storage. • Use webhooks to capture metadata and manifests: ◦ Subscribe to egress webhook events (egress_started / egress_ended) and store the metadata (egress ID, room, timestamps,

manifestLocation

) in your own database for later programmatic access. • Instrument your agent/app to persist session metadata: ◦ Log session start/end, participant lists, and references to recording locations to your analytics store so you can query by date range. • Use backup egress manifest if uploads fail: ◦ If a cloud upload fails and backup egress is enabled, you can retrieve backup files via the

manifestLocation

provided in the

egress_ended

webhook. If you want, I can help with any of these next steps: • Walk you through starting an egress command or API call and setting up S3/GCS destination. • Provide a short script to list egress jobs and download completed recordings from S3/GCS. • Suggest a minimal webhook schema to capture the fields you’ll want for time-range queries. • Troubleshoot why

lk egress list

is empty (if you think you did start egress): tell me which project/context you’re using and whether you started egress from an agent or from your app. Which of these would you like to do first? Sources: Streamline troubleshooting with Agent Observability | LiveKit | LiveKit Agent Observability | Debug and optimize voice agents | LiveKit | Universal Egress | LiveKit | Backup Egress Recordings, with interactive elements

I saw in the lk cli codebase experimental/disabled code for a lk records command, does it mean that it is in the roadmap?

Thanks — I’m responding to your question about the experimental/disabled

lk records

code you found and whether that means a

lk records

command is on the roadmap. Short answer: • I don’t have any confirmed information that

lk records

is an announced or guaranteed roadmap item. Experimental or disabled code in the CLI repo shows that the team has explored the idea, but it does not confirm an official release plan or timeline. What that means for you now: • The only supported, programmatic path to export recordings and session artifacts is Egress (start egress, upload to S3/GCS/Azure, or use backup manifests). The

lk

CLI currently exposes egress-related commands (

lk egress list

,

lk egress stop

) but no stable

lk records

command to fetch historical/dashboard-held recordings directly. • The Cloud dashboard still lets you view and download recordings/transcripts manually, but that data is not accessible programmatically from LiveKit servers without using egress. If you’d like, I can help with any of the following: • Walk through using Egress (API or agent) to export recordings to storage and then a script to download them. • Draft a simple webhook + storage workflow so you capture everything you need for time-range queries going forward. • Look at the experimental CLI code you found and explain what it appears to do (if you can paste the snippet or point me to the file). Which would you prefer? , with interactive elements

there's also a PR for supporting analytics, and it seem to work, how does it extract analytical/session data if it's not programmatically accessible?

Thanks — I’m replying about the analytics PR you mentioned and how the dashboard is getting that session/analytical data even though there’s no public CLI/API to fetch it. Short answer • The analytics feature in the dashboard is fed by LiveKit’s internal observability pipeline (agent observability). That pipeline gathers audio, transcripts, traces, and metadata and surfaces them in the Cloud UI. • Those internal observability endpoints are not exposed as a public API or stable CLI command today. Programmatic access to the same data still requires using Egress (to export artifacts) or collecting the data yourself via webhooks/agent instrumentation. What the analytics/observability pipeline does • It aggregates session telemetry (timestamps, participant events, trace spans), synchronized audio/transcript playback, and logs from agents and stores them for the dashboard. • The dashboard reads from that internal store to provide playback and analytic views. That store is not available as a public endpoint for arbitrary programmatic queries. Why you can see analytics in the dashboard but not via CLI/API • The dashboard uses internal services and pipelines that are not part of the public CLI/API surface. Showing data in the UI doesn’t imply a supported programmatic endpoint exists. • Any experimental CLI code or PRs you’ve seen may be exploring exposing parts of that functionality, but experimental or prototype code doesn’t guarantee a released CLI/API. I don’t have confirmation that those PRs will become a supported feature or when. Practical options to get programmatic access now • Export via Egress: ◦ Start egress to export recordings/transcripts to S3/GCS/Azure. Once exported, you can programmatically access the files from storage. • Capture events via webhooks: ◦ Subscribe to events (egress_started/egress_ended, participant_* events) and persist the metadata you need (timestamps, room IDs, egress manifestLocation) into your analytics DB for time-range queries. • Instrument your agent/app: ◦ Log session start/end, participant lists, metrics, and links to exported artifacts to a DB you control so you can query and analyze historically. • Use backup egress manifests: ◦ If uploads fail but backup egress is enabled, the

egress_ended

webhook includes a

manifestLocation

you can use to retrieve temporary copies. If you want me to dig into the PR you saw • Paste the PR link or the key code snippets and I’ll explain what it appears to do, which parts are experimental, and whether it looks like it’s merely UI-facing or actually hooks into an API we could rely on. • Tell me the exact analytics data you want programmatic access to (e.g., full recordings, per-participant audio, transcripts, trace spans, summary metrics). That helps me recommend the simplest supported path (egress vs webhooks vs instrumentation). Which analytics items do you need to access programmatically, and can you share the PR link or snippets you found? I’ll analyze it and recommend next steps. , with interactive elements

Thanks — I’m following up about the PR you shared (#784) and your question about how the analytics/dashboard data is being accessed. Short answer • The PR adds an experimental

lk records

command that reads from the same internal observability pipeline the Cloud dashboard uses. That pipeline collects audio, transcripts, traces, and metadata for Agent Observability. • The presence of an experimental/PR implementation does not mean there is a released, supported CLI/API to fetch dashboard-only data yet. The dashboard’s observability store isn’t exposed as a public programmatic endpoint today. What the PR appears to do • It surfaces session/record-like artifacts from the internal observability store so the CLI can show or download them in a developer workflow. • Because it relies on internal observability infrastructure, it’s marked experimental/disabled in the repo: it’s exploring functionality but is not a guaranteed, supported feature or part of the stable CLI today. Why you can see analytics in the dashboard but not retrieve them programmatically • The Cloud UI reads from LiveKit’s internal observability services that store telemetry and synchronized session artifacts for 30 days. • That internal store is not currently available via public CLI commands or APIs. Programmatic access to the same artifacts requires exporting them with Egress or capturing them yourself (webhooks, agent instrumentation). Supported ways to get analytics/recordings programmatically today • Egress: start egress (room/track/room composite) to export recordings/transcripts to your S3/GCS/Azure bucket; then access them via those storage APIs. • Webhooks: subscribe to egress and session events (egress_started / egress_ended, participant_*), persist metadata (timestamps, room IDs,

manifestLocation

) to your DB for time-range queries. • Agent/app instrumentation: log session start/end, participant info, metrics to your analytics backend. • Backup egress: if uploads fail and backup storage is used, the

egress_ended

webhook includes a

manifestLocation

you can use to download backup files. How I can help next • If you want, I can review the PR code in detail and explain exactly which internal endpoints it calls and which fields it extracts (paste any snippets or point to specific files). • I can recommend a minimal export/analytics workflow (webhook schema + storage layout) so you can reliably query sessions by date range. • I can provide a short script to: a. Start egress for new sessions, and b. List egress jobs and download completed recordings from S3/GCS. Which do you want me to do next — review the PR code, help set up an egress+storage workflow, or draft the webhook/DB schema for historical queries? Sources: Streamline troubleshooting with Agent Observability | LiveKit | LiveKit Agent Observability | Debug and optimize voice agents | LiveKit | examples/frontdesk | livekit/livekit_composite | DeepWiki | Agents (Python) | livekit/livekit_composite | DeepWiki | Backup Egress Recordings, with interactive elements