no operation provided:)

[sfz009900]

My.env configuration is as follows:

## LLM Providers
OPEN_AI_KEY=keyxxxxxxxxxxxxxxxxxxxxx
OPEN_AI_SERVER_URL=https://api.openai.com/v1

ANTHROPIC_API_KEY=
ANTHROPIC_SERVER_URL=https://api.openai.com/v1

## Custom LLM provider
LLM_SERVER_URL=https://api.deepseek.com
LLM_SERVER_KEY=xxxxxxxxxxxxxxx
LLM_SERVER_MODEL=deepseek-chat
LLM_SERVER_CONFIG_PATH=/opt/pentagi/conf/deepseek.provider.yml

Then my docker-compose.yml configuration is as follows:

  pentagi:
    image: vxcontrol/pentagi:latest
    restart: unless-stopped
    container_name: pentagi
    hostname: pentagi
    expose:
      - 8443/tcp
    ports:
      - 127.0.0.1:8443:8443
    depends_on:
      - pgvector
    environment:
      - DOCKER_GID=998
      - CORS_ORIGINS=${CORS_ORIGINS:-}
      - COOKIE_SIGNING_SALT=${COOKIE_SIGNING_SALT:-}
      - OPEN_AI_KEY=${OPEN_AI_KEY:-}
      - OPEN_AI_SERVER_URL=${OPEN_AI_SERVER_URL:-}
      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY:-}
      - ANTHROPIC_SERVER_URL=${ANTHROPIC_SERVER_URL:-}
      - LLM_SERVER_URL=${LLM_SERVER_URL:-}
      - LLM_SERVER_KEY=${LLM_SERVER_KEY:-}
      - LLM_SERVER_MODEL=${LLM_SERVER_MODEL:-}
      - LLM_SERVER_CONFIG_PATH=${LLM_SERVER_CONFIG_PATH:-}
      - PROXY_URL=${PROXY_URL:-}
      - SCRAPER_PUBLIC_URL=${SCRAPER_PUBLIC_URL:-}
      - SCRAPER_PRIVATE_URL=${SCRAPER_PRIVATE_URL:-}
      - PUBLIC_URL=${PUBLIC_URL:-}
      - STATIC_DIR=${STATIC_DIR:-}
      - STATIC_URL=${STATIC_URL:-}
      - SERVER_PORT=${SERVER_PORT:-8443}
      - SERVER_HOST=${SERVER_HOST:-0.0.0.0}
      - SERVER_SSL_CRT=${SERVER_SSL_CRT:-}
      - SERVER_SSL_KEY=${SERVER_SSL_KEY:-}
      - SERVER_USE_SSL=${SERVER_USE_SSL:-true}
      - OAUTH_GOOGLE_CLIENT_ID=${OAUTH_GOOGLE_CLIENT_ID:-}
      - OAUTH_GOOGLE_CLIENT_SECRET=${OAUTH_GOOGLE_CLIENT_SECRET:-}
      - OAUTH_GITHUB_CLIENT_ID=${OAUTH_GITHUB_CLIENT_ID:-}
      - OAUTH_GITHUB_CLIENT_SECRET=${OAUTH_GITHUB_CLIENT_SECRET:-}
      - DATABASE_URL=postgres://${PENTAGI_POSTGRES_USER:-postgres}:${PENTAGI_POSTGRES_PASSWORD:-postgres}@pgvector:5432/${PENTAGI_POSTGRES_DB:-pentagidb}?sslmode=disable
      - GOOGLE_API_KEY=${GOOGLE_API_KEY:-}
      - GOOGLE_CX_KEY=${GOOGLE_CX_KEY:-}
      - TRAVERSAAL_API_KEY=${TRAVERSAAL_API_KEY:-}
      - TAVILY_API_KEY=${TAVILY_API_KEY:-}
      - PERPLEXITY_API_KEY=${PERPLEXITY_API_KEY:-}
      - PERPLEXITY_MODEL=${PERPLEXITY_MODEL:-sonar}
      - PERPLEXITY_CONTEXT_SIZE=${PERPLEXITY_CONTEXT_SIZE:-low}
      - LANGFUSE_BASE_URL=${LANGFUSE_BASE_URL:-}
      - LANGFUSE_PROJECT_ID=${LANGFUSE_PROJECT_ID:-}
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY:-}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY:-}
      - OTEL_HOST=${OTEL_HOST:-}
      - DOCKER_HOST=${DOCKER_HOST:-unix:///var/run/docker.sock}
      - DOCKER_TLS_VERIFY=${DOCKER_TLS_VERIFY:-}
      - DOCKER_CERT_PATH=${DOCKER_CERT_PATH:-}
      - DOCKER_INSIDE=${DOCKER_INSIDE:-false}
      - DOCKER_SOCKET=${DOCKER_SOCKET:-}
      - DOCKER_NETWORK=${DOCKER_NETWORK:-}
      - DOCKER_PUBLIC_IP=${DOCKER_PUBLIC_IP:-}
      - DOCKER_WORK_DIR=${DOCKER_WORK_DIR:-}
      - DOCKER_DEFAULT_IMAGE=${DOCKER_DEFAULT_IMAGE:-}
    logging:
      options:
        max-size: 50m
        max-file: '7'
    volumes:
      - pentagi-data:/opt/pentagi/data
      - pentagi-ssl:/opt/pentagi/ssl
      - /var/run/docker.sock:/var/run/docker.sock
      - ./deepseek.provider.yml:/opt/pentagi/conf/deepseek.provider.yml
    user: root:root # while using docker.sock

But return visit https://localhost:8443/api/v1/graphql: {"errors":[{"message":"no operation provided","extensions":{"code":"GRAPHQL_VALIDATION_FAILED"}}],"data":null}

[sfz009900]

This is the result of "docker exec -it pentagi/ opt/pentagi/bin/ ctester-verbose ":

docker exec -it pentagi /opt/pentagi/bin/ctester -verbose
2025/03/28 12:34:16 Warning: Error loading .env file: open .env: no such file or directory
Testing Custom Provider with configuration:
=================================================
Testing agent: simple
Model: deepseek-chat
-------------------------------------------------

Test Results:

Basic Tests:
[✓] Completion: What is 2+2? Write only the... (14.242s)
[✓] Completion: Write 'Hello World' in uppe... (21.765s)
[✓] System-User: Count from 1 to 5, separate... (12.296s)
[✓] System-User: Calculate 5 * 10 and provid... (16.206s)
[✓] SimpleJSON: Return a JSON with a person... (14.824s)
[✓] Basic echo function (16.853s)

Advanced Tests:
[✓] JSON response function (13.754s)
[✓] Search query (12.433s)
[✗] Ask advice (77.670s)
    Error: model did not call a function, responded with text: The error **"cannot find package"** in Go devel...

Summary: 8/9 (88.89%) successful tests
Average latency: 22.227s

Testing agent: simple_json
Model: deepseek-chat
-------------------------------------------------

Test Results:

Basic Tests:
[✓] SimpleJSON: Return a JSON with a person... (25.337s)
[✓] SimpleJSON: Create a JSON object with f... (16.693s)
[✓] SimpleJSON: Generate a JSON response fo... (12.745s)
[✓] SimpleJSON: Create a JSON array of 3 co... (16.171s)

Advanced Tests:
[✓] SimpleJSON: What is 2+2? Explain your a... (20.080s)

Summary: 5/5 (100.00%) successful tests
Average latency: 18.205s

Testing agent: agent
Model: deepseek-chat
-------------------------------------------------

Test Results:

Basic Tests:
[✓] Completion: What is 2+2? Write only the... (26.768s)
[✓] Completion: Write 'Hello World' in uppe... (16.566s)
[✓] System-User: Count from 1 to 5, separate... (14.831s)
[✓] System-User: Calculate 5 * 10 and provid... (10.406s)
[✓] SimpleJSON: Return a JSON with a person... (18.705s)
[✓] Basic echo function (14.740s)

Advanced Tests:
[✓] JSON response function (11.463s)
[✓] Search query (46.328s)
[✗] Ask advice (75.559s)
    Error: model did not call a function, responded with text: The error `cannot find package` in Go developme...

Summary: 8/9 (88.89%) successful tests

But an error is still reported:

2025-03-28 20:39:47 time="2025-03-28T12:39:47Z" level=error msg="graphql request handled with errors" component=pentagi-gql duration="84.808µs" gql.errors="input: createFlow model provider is required\n" operation_name=createFlow operation_type=mutation
2025-03-28 20:39:47 [GIN] 2025/03/28 - 12:39:47 | 200 |     977.595µs |      172.22.0.1 | POST     "/api/v1/graphql"

[asdek]

Hello @sfz009900

Thank you for bringing this to our attention.

It appears there's a bug in the latest update where the LLM Provider isn't automatically populated when you first open the interface. To resolve this issue for now, please manually select the appropriate provider from the dropdown list.

I'm working on a fix to address this problem and will provide an update soon within this issue.

[sfz009900]

@asdek Same thing happened to me now 😅#19 (comment)

[asdek]

@sfz009900 I'll do my best 😉

[sfz009900]

Wow @asdek, you're a lifesaver! It'd be awesome if we could customize things though, like with an ollama embedding query or something. 😉

[sfz009900]

Hey @asdek sorry! Forgot I have a third-party API that's compatible with OpenAI. I can set it up for just OpenAI, and then the embedded model should be good to go :)

[sfz009900]

Hey @asdek, something seems off here... 🤔

[sfz009900]

It keeps trying to download "https://http.kali.org/kali/pool/main/b/bind9/dnsutils_9.18.16-1_all.deb" but that thing doesn't exist! It's stuck in a loop, lol.

[asdek]

@sfz009900

First of all, thank you so much for your kind words! I'm glad the solution was helpful to you. 😊

---

Regarding Customizing Embeddings with Ollama:

"It'd be awesome if we could customize things though, like with an Ollama embedding query or something"

Support for Ollama was implemented in the early versions of the product. However, as the number of functions grew and with the increasing need for strict adherence to JSON structures when invoking functions, we had to discontinue that support. Unfortunately, I couldn't find models that were sophisticated enough to handle complex tasks like attack simulations. These LLMs also tended to deviate from the required function structure, making it challenging to guide them back on track. The context length likely plays a significant role in this issue.

---

Using an OpenAI-Compatible Backend:

"Forgot I have a third-party API that's compatible with OpenAI. I can set it up for just OpenAI, and then the embedded model should be good to go :)"

If you have an OpenAI-compatible backend, the Custom provider is designed to work seamlessly with such interfaces. You can finely tune the model's parameters through your own configuration file to make it respond more accurately to your requests. All parameters specified in the OpenAI API specification are supported.

You can find more details here: Custom Provider Configuration.

---

Issue with Installing dnsutils:

"Something seems off here..."

It keeps trying to download "https://http.kali.org/kali/pool/main/b/bind9/dnsutils_9.18.16-1_all.deb" but that thing doesn't exist! It's stuck in a loop, lol."

Regarding the issue you're experiencing with the installation of the dnsutils package, it seems there's a problem with the apt repository:

• Apt Update Error: The error during apt update indicates that the package lists aren't being updated due to a GPG key error. This means apt isn't aware of the latest packages, which is why it tries to download a package version that no longer exists on the server.

• Cached Package Lists: It's possible that when the Docker image was built, the package lists were cached and haven't been updated since. Therefore, apt is attempting to fetch an outdated version of dnsutils

For Now:

You can explicitly specify in your initial task prompt which Docker image the agent should use, and it will follow those instructions. If you find a more suitable image from another repository, please let me know! 😉

---

About the Embedding Server Support:

I'll do my best to deliver this update tomorrow. However, today's update includes many useful fixes, including resolving the issue you mentioned earlier with the empty provider value in the UI and more.

---

Additional Information on Model Behavior:

PentAGI includes a separate agent called "reflector" that monitors the format of responses and attempts to guide the model when it makes mistakes. However, with smaller models, this doesn't always help, and tasks sometimes still end with errors. For example, I couldn't get Mistral-3.1-24B to work properly, and there were issues with Qwen-32B as well.

Configuring the initial prompt to ensure models strictly follow instructions often requires significant effort. Here's an example of one of the prompts for testing web applications, where we have to provide detailed instructions on how to perform and decompose tasks: Web Pentest Base Prompt.