 prathameshmudgale on  ~/Work/agent-starter-python feat/kalpataru # lk agent list Using project [testproject] No agents found  prathameshmudgale on  ~/Work/agent-starter-python feat/kalpataru # lk project list ┌───────────────────────┬──────────────────────────────────────────────────┬─────────────────┐ │ Name │ URL │ API Key │ ├───────────────────────┼──────────────────────────────────────────────────┼─────────────────┤ │ testproject │ wss://testproject-u43es40w.livekit.cloud │ API2ixg8VTPJSUj │ │ qualif-inbound-prod │ wss://qualif-inbound-prod-nf47x7fn.livekit.cloud │ APIfUPxZR7qgAPC │ │ * test_project │ wss://testproject-u43es40w.livekit.cloud │ APIusFLLYSw8ZZy │ Even though I switched to test_project when I do lk agent list it uses testproject

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

When I try to run the agent locally and then connect to it via playground I am not able to uv run src/agent.py dev 2025-12-17 154303,048 - DEBUG asyncio - Using selector: KqueueSelector 2025-12-17 154303,049 - DEV livekit.agents - Watching /Users/prathameshmudgale/Work/agent-starter-python/src 2025-12-17 154304,380 - DEBUG asyncio - Using selector: KqueueSelector 2025-12-17 154304,386 - INFO livekit.agents - starting worker {"version": "1.2.18", "rtc-version": "1.0.12"} 2025-12-17 154304,386 - INFO livekit.agents - starting inference executor 2025-12-17 154304,393 - INFO livekit.agents - initializing process {"pid": 38464, "inference": true} 2025-12-17 154305,694 - DEBUG livekit.agents - initializing inference runner {"runner": "lk_end_of_utterance_multilingual", "pid": 38464, "inference": true} None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2025-12-17 154307,929 - DEBUG livekit.agents - inference runner initialized {"runner": "lk_end_of_utterance_multilingual", "elapsed_time": 2.234338874986861, "pid": 38464, "inference": true} 2025-12-17 154307,931 - INFO livekit.agents - process initialized {"pid": 38464, "inference": true, "elapsed_time": 3.54} 2025-12-17 154307,930 - DEBUG asyncio - Using selector: KqueueSelector {"pid": 38464, "inference": true} 2025-12-17 154308,179 - INFO livekit.agents - registered worker {"id": "AW_ArZ6KNuC2Gv7", "url": "wss://testproject-u43es40w.livekit.cloud", "region": "India South", "protocol": 16}

Thanks — I’m responding to your issue running the agent locally and not being able to connect from the Agents Playground. Quick findings from your logs: the agent started and registered with LiveKit Cloud (worker id AW_ArZ6KNuC2Gv7, URL wss://testproject-u43es40w.livekit.cloud). The warning about missing PyTorch/TensorFlow/Flax only affects model availability, not registration or connectivity. Please try these checks and tell me any errors you see in the Playground (or paste them): 1. Confirm CLI/project auth ◦ Run

lk cloud auth

and sign in, then confirm the Playground is using the same LiveKit account/project. 1. Force the project when testing ◦ From your agent directory run:

lk agent list --project test_project

to ensure the CLI/project context matches the agent’s registered project. 1. Environment variables / .env ◦ Ensure

LIVEKIT_API_KEY

,

LIVEKIT_API_SECRET

, and

LIVEKIT_URL

in your

.env

match the project shown in your agent logs (wss://testproject-u43es40w.livekit.cloud). 1. Download required files (if you need model-based features) ◦ Run:

uv run src/agent.py download-files

then restart

uv run src/agent.py dev

. Missing ML libs means model functionality is limited until models are downloaded and supported. 1. Network / WebSocket reachability ◦ Verify your machine can reach the LiveKit WebSocket URL (no corporate firewall blocking):

curl -v <https://testproject-u43es40w.livekit.cloud>

and check browser console/network tab when the Playground attempts to connect for WebSocket errors. 1. Playground errors / logs ◦ Open the browser devtools Console/Network when connecting from the Agents Playground and paste any error messages (auth failure, 401, WebSocket close codes, CORS, etc.). Specific error text will pinpoint the problem. If you prefer, paste: • the Playground error message(s), • your

.env

(redact secrets), • and the output of

lk agent list --project test_project

. I’ll use that to give the exact next step. Sources: Voice AI quickstart | LiveKit docs | Agents (Python) | livekit/livekit_composite | DeepWiki | LiveKit CLI Setup | LiveKit Docs | Rust SDKs | livekit/livekit_composite | DeepWiki | Diagnosing Connection Errors with Connection Test Utility

there is no error message on playground, only "Waiting for agent audio track…" LIVEKIT_URL=wss://testproject-********.livekit.cloud LIVEKIT_API_KEY=********* LIVEKIT_API_SECRET=********

Thanks — you're seeing "Waiting for agent audio track…" which means the Playground connected but no audio track was published. From your logs the agent registered successfully. Let's check the most likely causes. Please try these steps and share results/errors you see: 1. Confirm the session actually starts • In your agent code, ensure you call

await session.start(room=ctx.room, agent=...)

. If you can, paste the small section where you create/start the AgentSession. 1. Check audio output is enabled • If you set room options, ensure

audio_output=True

. If using an avatar worker, audio may be handled differently. 1. Verify the agent was dispatched and started a job • Run:

lk agent status
  lk agent logs

Share any lines showing

AgentSession started

,

published audio track

, or exceptions. 1. Test TTS/console locally • Try

uv run src/agent.py console

(or a minimal example that calls

session.start()

and

session.generate_reply(...)

) to confirm TTS/LLM will produce audio locally. 1. Confirm env and token/project match • You already redacted secrets — confirm

LIVEKIT_URL

exactly matches the URL in the agent logs and the Playground project. (No action if they already match.) 1. Browser DevTools • In the Playground, open DevTools → Console & Network. Look for WebSocket events or log lines when the room joins and when tracks are published; paste any relevant errors or warnings. If you prefer paste here: • The AgentSession start code (redact secrets), • Output of

lk agent status

and recent

lk agent logs

. I’ll analyze those and tell you the exact next step. Sources: Voice AI quickstart | LiveKit docs | Virtual avatar plugins | LiveKit Docs | Agent session | LiveKit docs | Deploying to custom environments | LiveKit docs

when I run " uv run src/agent.py console" it runs perfectly fine

Great — since console mode works, the agent's LLM/TTS is fine. That narrows the problem to room dispatch/publishing when used via the Playground. Please provide/do the following so I can pinpoint the issue: 1. Paste the AgentSession start snippet (redact secrets). I’m checking you call:

await session.start(room=ctx.room, agent=...)

1. Confirm the agent name used in your rtc handler or dispatch, and how you generate the room token in Playground. If you dispatch by agent name, the name must match the token/dispatch rule. 2. Share these CLI outputs (paste here): •

lk agent status

•

lk agent logs

(last ~100 lines) 1. While Playground is connected, open browser DevTools → Console/Network and paste any WebSocket events or console logs around room join/publish time. 2. Quick check you can run now: • From the machine running the agent, watch for a publish event in logs while Playground is connected. Run

lk agent logs --follow

and then connect via Playground; paste any new lines about

published audio track

or errors. Send the code snippet and those outputs and I’ll tell you the exact change to make.

lk agent status Using project [testproject] Using agent [AW_ArZ6KNuC2Gv7] no agents found even though the agent is running locally

You're seeing

no agents found

because the CLI is querying the project currently selected for the command (it shows "Using project [testproject]") while your local worker registered under the other project (

test_project

/ the URL wss://testproject-u43es40w.livekit.cloud). The agent is running but the CLI is looking in a different project. Do these checks and paste the results: 1. Explicitly list agents for the project the agent registered with:

lk agent list --project test_project

1. Confirm the CLI project selection and any local override: ◦ Run

lk project list

and verify which project has the

*

. ◦ Check for a

livekit.toml

in your agent directory that may be overriding the default:

cat livekit.toml

1. Show agent registration from the agent logs (while agent is running):

lk agent logs --follow

Paste the lines that mention registration (the lines showing the worker id and URL). 1. If

lk agent list --project test_project

still shows no agents, run:

lk agent list --project test_project --verbose

and paste that output. Once you paste the outputs (especially the

lk agent list --project test_project

result and the

livekit.toml

contents), I’ll tell you the exact change to make so `lk agent status`/`lk agent list` show your running agent and so the Playground can dispatch to it.

lk agent list --project test_project Using project [testproject] No agents found

Thanks — I’m following up on why

lk agent list --project test_project

still shows “Using project [testproject]” and “No agents found.” Two important points: • The CLI’s

--project

and

lk project set

expect the project identifier/subdomain (e.g.,

testproject-u43es40w

), not the friendly name with underscores (

test_project

). • A

livekit.toml

in your current directory can override the CLI default project. Please run these commands and paste the outputs: 1. Show the exact projects and which is set:

lk project list

1. Try listing agents with the project subdomain your agent registered to (use the subdomain shown in your agent logs, e.g.

testproject-u43es40w

):

lk agent list --project testproject-u43es40w

1. Check for a local override file:

cat livekit.toml

(if the file doesn’t exist, say “no file”) If the agent then appears with

lk agent list --project <subdomain>

, run:

lk project set testproject-u43es40w
lk agent status

and confirm the agent shows as RUNNING. Paste the outputs of steps 1–3 and I’ll tell you the exact edit or command to make the CLI and Playground use the same project so dispatch/publishing works.

[project] subdomain = "testproject-u43es40w" [agent] id = "AW_ArZ6KNuC2Gv7"

Thanks — I can see your

livekit.toml

points to the correct subdomain and agent id. Do this now so the CLI and Playground use the same project and we can confirm the agent is visible: 1. Set the project to the subdomain:

lk project set testproject-u43es40w

1. Verify agent visibility:

lk agent list
lk agent status

Paste the outputs here. If the agent appears, reconnect the Playground and watch

lk agent logs --follow

— paste any new lines about

AgentSession started

or

published audio track

. If the agent still doesn't appear, paste the outputs from steps 1–2 and your

lk agent logs --follow

output and I’ll tell you the exact next step.

no agents found

Thanks — you're right to be frustrated. The agent is running locally and registered, but the CLI still isn't showing it because the CLI commands are using a different project context or credentials. Let's verify and align everything step-by-step so the agent becomes visible and the Playground can dispatch to it. 1) Confirm the agent actually registered to the subdomain shown in the agent logs • While your agent is running, paste the registration lines from your agent logs (they look like the one you shared earlier). I need the exact subdomain and worker id lines, e.g.: ◦ registered worker {"id":"AW_...", "url":"wss://testproject-u43es40w.livekit.cloud", ...} 2) Re-auth and ensure the CLI uses the same project credentials • Run:

lk cloud auth

Sign in with the same LiveKit account you used for the Playground. 3) Explicitly list agents using the agent subdomain (exact value from logs) • Replace <subdomain> with the subdomain from the agent logs (e.g.

testproject-u43es40w

):

lk agent list --project <subdomain>
  lk agent list --project <subdomain> --verbose

4) Confirm your local config (.env and livekit.toml) • Paste the contents (redact API secrets) of: ◦

livekit.toml

(you already showed it; confirm its location: is it in ~/Work/agent-starter-python?) ◦

.env

(show

LIVEKIT_URL

value and confirm it matches the agent logs' URL) • If

LIVEKIT_URL

differs from the agent logs' URL, update it to match the subdomain the agent registered to. 5) If

lk agent list --project <subdomain>

still shows "No agents found" • Run

lk agent logs --follow

while you connect the Playground and paste any new log lines that appear. • Run

lk project list

and paste its output so I can confirm which project is currently set by the CLI. What I need from you now • The exact registration line(s) from your running agent logs (the lines showing the worker id and URL). • Output of: ◦

lk agent list --project testproject-u43es40w --verbose

◦

lk project list

◦ Confirmation of where

livekit.toml

is located and your `.env`'s

LIVEKIT_URL

value (redact secrets). Once you paste those, I’ll tell you the exact command or file edit to make the CLI show your running agent and get the Playground dispatching audio.

load_dotenv(".env.local") lk agent list --project testproject-u43es40w --verbose Using project [testproject] 2025-12-17T160902.092+0530 DEBUG lk config/livekit.go:89 loading livekit.toml file No agents found

Thanks — I can tell this is frustrating. You’ve confirmed the CLI is loading your local livekit.toml but still shows “No agents found” even though the agent log shows it registered. Let’s verify the CLI, credentials, and project alignment precisely. Please run these commands (paste outputs here; redact secrets): 1) Confirm which project the CLI thinks is active:

lk project list

2) Re-authenticate the CLI so it uses the same LiveKit account as the Playground:

lk cloud auth

(Complete the login flow, then re-run

lk project list

.) 3) Explicitly list agents against the subdomain while forcing the config file load:

lk agent list --project testproject-u43es40w --config ./livekit.toml --verbose

4) Show the agent registration lines from the running agent logs (paste the exact lines):

lk agent logs --follow

Look for lines like

registered worker {"id": "...", "url": "wss://<subdomain>.livekit.cloud", ...}

and paste those lines. 5) Confirm your .env values (redact secrets) — I only need the

LIVEKIT_URL

value and to know which env file you loaded: • Which file did you load? (

.env

,

.env.local

, etc.) • Value of

LIVEKIT_URL

(redact token parts), e.g.

<wss://testproject-u43es40w.livekit.cloud>

Why each step: •

lk project list

verifies CLI project selection. •

lk cloud auth

ensures CLI credentials match the Playground account. •

--config ./livekit.toml

forces the CLI to use the same local config the agent is using. • Agent log lines confirm the exact subdomain the agent registered to. • .env check confirms the agent and CLI are pointing to the same LiveKit URL. After you paste the requested outputs I’ll tell you the single exact command or file edit to make the agent appear in

lk agent list

and for the Playground to dispatch audio.

2025-12-17 154308,179 - INFO livekit.agents - registered worker {"id": "AW_ArZ6KNuC2Gv7", "url": "wss://testproject-u43es40w.livekit.cloud", "region": "India South", "protocol": 16}

Thanks — your agent log confirms it registered to wss://testproject-u43es40w.livekit.cloud. Next steps — these will align CLI, agent, and Playground so the agent appears in

lk

and can be dispatched: 1. Set the CLI project to the agent's subdomain:

lk project set testproject-u43es40w

1. Re-authenticate the CLI to ensure it uses the same account as the Playground:

lk cloud auth

Complete the login flow in the browser. 1. Ensure your agent process and CLI point at the same config: • Confirm

LIVEKIT_URL

in the env file you load (you said

load_dotenv(".env.local")

) equals:

<wss://testproject-u43es40w.livekit.cloud>

• Confirm

livekit.toml

in your agent directory has:

[project]
  subdomain = "testproject-u43es40w"

1. Now verify agent visibility:

lk agent list
lk agent status

1. While Playground is connected, follow agent logs to watch for session start / publish events:

lk agent logs --follow

Look for lines like

AgentSession started

and

published audio track

. If after step 1–4 the CLI still shows “No agents found,” paste these exact outputs: •

lk project list

•

lk agent list --verbose

• The

LIVEKIT_URL

value from

.env.local

(redact secrets) I’ll tell you the single change needed based on those outputs.