NVIDIA / NVIDIA/NemoClaw

[Ubuntu 22.04/26 GPU][DCode][Ollama] non-interactive response emits tool schema instead of requested plain text

Open
#11,937 0 comments 0 reactions 0 assignees View on GitHub
integration: dcode needs: triage NV QA platform: ubuntu
Dominant language
TypeScript
Stars
22.5k
Forks
3.1k
Avg merge
1d 1h
Merged PRs (30d)
715

Description

## Description

After a clean NemoClaw v0.0.126 install and successful Deep Agents Code onboarding to a reachable host Ollama service, a non-interactive DCode prompt that requests one exact word returns a serialized tool schema/tool-call payload instead of the requested plain-text response.

Platform scope: Independently reproduced on Ubuntu 22.04 and Ubuntu 26 GPU.

Regression: Unknown.

OpenShell issue: No evidence. The sandbox, route, onboarding, and host Ollama preflight are healthy.

## Environment

```text
NemoClaw: v0.0.126
Agent: Deep Agents Code (DCode app v0.1.55)
Provider: host Ollama
Model: llama3.2:1b
Platforms: Ubuntu 22.04 x86_64 and Ubuntu 26 GPU x86_64
```

## Steps to Reproduce

1. Install NemoClaw v0.0.126.
2. Start a reachable host Ollama service with `llama3.2:1b` available.
3. Onboard a Deep Agents Code sandbox non-interactively using the host Ollama loopback route.
4. Confirm the sandbox is Ready and the inference route is healthy.
5. Run DCode non-interactively with a prompt that asks it to return exactly `OLLAMA`.

## Expected Result

DCode returns exactly `OLLAMA` as ordinary assistant text.

## Actual Result

DCode exits successfully but emits a serialized function/tool schema or tool-call payload, such as a `read_file` function definition, rather than `OLLAMA`.

## Reproducibility

- Ubuntu 22.04: reproduced in two clean targeted runs.
- Ubuntu 26 GPU: reproduced in an independent targeted run.
- Each run reached product assertions with two checks passing and one failing; no setup/teardown hook or environment blocker was present.

## Related Issues

- #8121 covers an opaque managed non-interactive error against an NVIDIA endpoint.
- #11806 covers a remote agent failure.

Neither covers successful host-Ollama inference returning a serialized tool schema instead of plain text.

Contributor guide

Open the contributing guide

Research direction

Start by reproducing the non-interactive DCode prompt against the host Ollama service with llama3.2:1b, then trace the response handling from the DCode invocation through the inference route. Compare the serialized tool schema output with the expected plain-text response; done means the command returns exactly OLLAMA on the reported Ubuntu environments.

Written by the indexing model from the issue text.

Assessment

Tech stack
ollama, typescript
Domain
ai, backend-api-design
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
50/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.