Error adding custom provider (perplexity) "unsupported operand type(s) for +=: 'int' and 'str'"

[dintho]

Hi, i am having some issues with adding in perplexity AI as a custom provider

error from ui

unsupported operand type(s) for +=: 'int' and 'str'

deployment

docker container

]# docker exec -it aa5a6b10ffe7 /bin/bash
llms@aa5a6b10ffe7:/app$ llms
usage: llms [-h] [--config FILE] [--providers FILE] [-m MODEL] [--chat REQUEST] [-s PROMPT] [--tools TOOLS] [--image IMAGE] [--audio AUDIO] [--file FILE] [--out MODALITY] [--args PARAMS] [--raw] [--nohistory] [--nostore] [--list] [--check PROVIDER]
            [--serve PORT] [--enable PROVIDER] [--disable PROVIDER] [--default MODEL] [--init] [--update-providers] [--reset [TYPE]] [--auth EXTENSION] [--logprefix PREFIX] [--verbose] [--add [EXTENSION]] [--remove [EXTENSION]] [--update [EXTENSION]]
            [--adduser USERNAME] [--removeuser USERNAME] [--listusers] [--lockuser [USERNAME]] [--unlockuser USERNAME]

llms v3.0.42

configs

i added this to llms.json

    "perplexity": {
      "enabled": true,
      "id": "perplexity",
      "npm": "@ai-sdk/openai-compatible",
      "api": "https://api.perplexity.ai",
      "no_tools": true,
      "env": [
        "PERPLEXITY_API_KEY"
      ],
      "models": {
        "sonar": {
          "id": "sonar",
          "name": "Perplexity Sonar"
        },
        "sonar-pro": {
          "id": "sonar-pro",
          "name": "Perplexity Sonar Pro"
        },
        "sonar-deep-research": {
          "id": "sonar-deep-research",
          "name": "Perplexity Sonar Deep Research"
        }
      }
    }

Errorr logs

2026-04-08 06:22:43.604289+00:00======== Running on http://0.0.0.0:30335 ========
2026-04-08 06:22:43.604330+00:00(Press CTRL+C to quit)
2026-04-08 06:23:24.107246+00:00provider: perplexity OpenAiCompatible
2026-04-08 06:23:24.109400+00:00POST https://api.perplexity.ai/chat/completions
2026-04-08 06:23:24.109583+00:00{
2026-04-08 06:23:24.109631+00:00"model": "sonar",
2026-04-08 06:23:24.109655+00:00"messages": [
2026-04-08 06:23:24.109678+00:00{
2026-04-08 06:23:24.109701+00:00"timestamp": 1775628830151,
2026-04-08 06:23:24.109743+00:00"role": "user",
2026-04-08 06:23:24.109766+00:00"content": [
2026-04-08 06:23:24.109789+00:00{
2026-04-08 06:23:24.109812+00:00"type": "text",
2026-04-08 06:23:24.109836+00:00"text": "hi"
2026-04-08 06:23:24.109860+00:00}
2026-04-08 06:23:24.109901+00:00],
2026-04-08 06:23:24.109929+00:00"model": "Perplexity Sonar",
2026-04-08 06:23:24.109953+00:00"usage": {
2026-04-08 06:23:24.109976+00:00"tokens": 2,
2026-04-08 06:23:24.109999+00:00"price": 0,
2026-04-08 06:23:24.110023+00:00"cost": 0
2026-04-08 06:23:24.110062+00:00}
2026-04-08 06:23:24.110085+00:00}
2026-04-08 06:23:24.110108+00:00],
2026-04-08 06:23:24.110130+00:00"modalities": [
2026-04-08 06:23:24.110154+00:00"text"
2026-04-08 06:23:24.110176+00:00],
2026-04-08 06:23:24.110236+00:00"metadata": {
2026-04-08 06:23:24.110259+00:00"tools": "none"
2026-04-08 06:23:24.110282+00:00},
2026-04-08 06:23:24.110304+00:00"stream": false
2026-04-08 06:23:24.110327+00:00}
2026-04-08 06:23:25.366572+00:00ERROR: Provider perplexity failed: unsupported operand type(s) for +=: 'int' and 'str'
2026-04-08 06:23:25.367530+00:00Traceback (most recent call last):
2026-04-08 06:23:25.367567+00:00File "/usr/local/lib/python3.11/site-packages/llms/main.py", line 1987, in g_chat_completion
2026-04-08 06:23:25.367591+00:00await filter_func(final_response, context)
2026-04-08 06:23:25.367614+00:00File "/usr/local/lib/python3.11/site-packages/llms/extensions/app/__init__.py", line 921, in chat_response
2026-04-08 06:23:25.367694+00:00total_costs += request.get("cost", 0) or 0
2026-04-08 06:23:25.367722+00:00TypeError: unsupported operand type(s) for +=: 'int' and 'str'
2026-04-08 06:23:25.367746+00:002026-04-08T06:23:25.367746180Z
2026-04-08 06:23:25.371193+00:00ERROR [app]: run_chat unsupported operand type(s) for +=: 'int' and 'str'
2026-04-08 06:23:25.371253+00:00Traceback (most recent call last):
2026-04-08 06:23:25.371277+00:00File "/usr/local/lib/python3.11/site-packages/llms/extensions/app/__init__.py", line 229, in run_chat
2026-04-08 06:23:25.371301+00:00await ctx.chat_completion(chat_req, context=context_req)
2026-04-08 06:23:25.371359+00:00File "/usr/local/lib/python3.11/site-packages/llms/main.py", line 3391, in chat_completion
2026-04-08 06:23:25.371383+00:00return await self.app.chat_completion(chat, context=context)
2026-04-08 06:23:25.371421+00:00^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-04-08 06:23:25.371445+00:00File "/usr/local/lib/python3.11/site-packages/llms/main.py", line 3069, in chat_completion
2026-04-08 06:23:25.371484+00:00response = await g_chat_completion(chat, context)
2026-04-08 06:23:25.371508+00:00^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-04-08 06:23:25.371531+00:00File "/usr/local/lib/python3.11/site-packages/llms/main.py", line 2004, in g_chat_completion
2026-04-08 06:23:25.371621+00:00raise first_exception
2026-04-08 06:23:25.371655+00:00File "/usr/local/lib/python3.11/site-packages/llms/main.py", line 1987, in g_chat_completion
2026-04-08 06:23:25.371690+00:00await filter_func(final_response, context)
2026-04-08 06:23:25.371724+00:00File "/usr/local/lib/python3.11/site-packages/llms/extensions/app/__init__.py", line 921, in chat_response
2026-04-08 06:23:25.371823+00:00total_costs += request.get("cost", 0) or 0
2026-04-08 06:23:25.371860+00:00TypeError: unsupported operand type(s) for +=: 'int' and 'str'
2026-04-08 06:23:25.371894+00:002026-04-08T06:23:25.371894194Z

Troubleshooting steps ive taken

• ive tried setting the costs manually in the json file
• tried moving the config to providers-extra.json
• i went to cluade.al for help

what the AI said to do:

modify /usr/local/lib/python3.11/site-packages/llms/extensions/app/__init__.py or llms/llms/extensions/app/__init__.py lines 921-923

-                total_costs += request.get("cost", 0) or 0
-                total_input += request.get("inputTokens", 0) or 0
-                total_output += request.get("outputTokens", 0) or 0
+                try:
+                    _c = request.get("cost", 0) or 0
+                    if isinstance(_c, dict): total_costs += float(_c.get("total_cost", 0))
+                    elif isinstance(_c, str) and _c.startswith("{"): total_costs += float(__import__("json").loads(_c).get("total_cost", 0))
+                    elif _c: total_costs += float(_c)
+                except Exception: pass
+                total_input += int(float(request.get("inputTokens", 0) or 0))
+                total_output += int(float(request.get("outputTokens", 0) or 0))

Results

this did get rid of the error, however i dont know the code well enough to make PR and just dump AI code as a fix.
i did patch my local docker container and it seems fine but i dont know how that section of code is used by other models

comment

i am really new to this project and went digging through docs, and nothing really fit what if the model is in https://models.dev but not already listed in providers.json so i am totally open to the possibility that i did this adding of perplexity wrong or that im missing something

[mythz]

Not sure where you got your json from but the definition from models.dev for the provider that uses https://api.perplexity.ai was:

{
    "perplexity-agent": {
        "id": "perplexity-agent",
        "env": [
            "PERPLEXITY_API_KEY"
        ],
        "npm": "@ai-sdk/openai",
        "api": "https://api.perplexity.ai/v1",
        "name": "Perplexity Agent",
        "doc": "https://docs.perplexity.ai/docs/agent-api/models",
        "models": {
            "perplexity/sonar": {
                "id": "perplexity/sonar",
                "name": "Sonar",
                "family": "sonar",
                "attachment": false,
                "reasoning": false,
                "tool_call": true,
                "temperature": true,
                "knowledge": "2025-09-01",
                "release_date": "2024-01-01",
                "last_updated": "2025-09-01",
                "modalities": {
                    "input": [
                        "text"
                    ],
                    "output": [
                        "text"
                    ]
                },
                "open_weights": false,
                "cost": {
                    "input": 0.25,
                    "output": 2.5,
                    "cache_read": 0.0625
                },
                "limit": {
                    "context": 128000,
                    "output": 8192
                }
            }
        }
    }
}

The @ai-sdk/openai npm says to use an OpenAI specific provider in openai.py not the Open AI general compatible provider.

Given that you should be able to inherit the definition in models.dev with the name of the provider and enable it with:

"perplexity-agent": {
    "enabled": true
},

Now I'm not sure it will work as there's no free trial for Perplexity and the minimum credits is $50 which I'm not paying.

Either way any customizations to support Perplexity would need to be added to the OpenAiProvider provider in openai.py

[dintho]

hey , yeh where i got my json was by hand trying to get it too enable i was trying regular perplexity not perpexity-agent

so i tried just adding ur suggested code

"perplexity-agent": {
"enabled": true
},

it does not enable when i try to enable it through the ui i get a Failed to enable perplexity-agent: HTTP 500 Internal Server Error
and the logs

2026-04-08 15:40:04.007592+00:00Provider perplexity-agent is missing 'npm' sdk
2026-04-08 15:40:04.012311+00:00Error handling request from <ip address of system>
2026-04-08 15:40:04.012361+00:00Traceback (most recent call last):
2026-04-08 15:40:04.012418+00:00File "/usr/local/lib/python3.11/site-packages/aiohttp/web_protocol.py", line 510, in _handle_request
2026-04-08 15:40:04.012443+00:00resp = await request_handler(request)
2026-04-08 15:40:04.012465+00:00^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2026-04-08 15:40:04.012488+00:00File "/usr/local/lib/python3.11/site-packages/aiohttp/web_app.py", line 569, in _handle
2026-04-08 15:40:04.012526+00:00return await handler(request)
2026-04-08 15:40:04.012549+00:00^^^^^^^^^^^^^^^^^^^^^^
2026-04-08 15:40:04.012572+00:00File "/usr/local/lib/python3.11/site-packages/llms/main.py", line 4300, in provider_handler
2026-04-08 15:40:04.012594+00:00provider_config, msg = enable_provider(provider)
2026-04-08 15:40:04.012631+00:00^^^^^^^^^^^^^^^^^^^^^^^^^
2026-04-08 15:40:04.012654+00:00File "/usr/local/lib/python3.11/site-packages/llms/main.py", line 2377, in enable_provider
2026-04-08 15:40:04.012677+00:00msg = provider.validate(**constructor_kwargs)
2026-04-08 15:40:04.012699+00:00^^^^^^^^^^^^^^^^^
2026-04-08 15:40:04.012735+00:00AttributeError: 'NoneType' object has no attribute 'validate'

chasing down the errors i eventually got a error
AttributeError: 'OpenAiProvider' object has no attribute 'load_models'. Did you mean: 'map_models'? and then i just made a full model stanza all by hand using the docs

if i copy in the full stanza you provided im still seeing the cost errors
There is also a 404 error in the logs that goes away when i use the api url from the perplexity docs (my api url provided in the initial issue)

[mythz]

Can't help from here as I'm not able to test it, but basically every time I add a provider I typically have to modify the provider class to add any provider-specific support for it. Not able to help here since I'm not able to test it, hopefully someone with a premium Perplexity account will be able to add any changes required to support it.