12 replies

Ollama and Kimi K2.5 cloud

Ask it these questions, in this order:

1. OpenClaw's Ollama provider is connecting to Ollama Cloud (ollama.com) using a thinking model (kimi-k2.5). The model returns content in the

thinking

thinking

field with an empty

response

response

field during streaming. The gateway hangs for 600 seconds then throws

FailoverError: LLM request timed out

FailoverError: LLM request timed out

. Is there a config setting to handle thinking model output, or does the Ollama provider only read the

response

response

field?
2. What is the correct

agents.list

agents.list

config to add fallback models to a per-agent primary override? Specifically: Agent Odin and Agent Mimir have

"model": {"primary": "openai-codex/gpt-5.4"}

"model": {"primary": "openai-codex/gpt-5.4"}

in their agents.list entries but no fallbacks. Does adding

"fallbacks": ["anthropic/claude-haiku-4-5"]

"fallbacks": ["anthropic/claude-haiku-4-5"]

to that same model object work, or does the per-agent primary need a different structure to inherit or define fallbacks?
3. Is there a configurable request timeout for LLM calls shorter than the default 600 seconds? The 10-minute hang before failover is too long for interactive Discord use.
4. Does OpenClaw auto-regenerate per-agent

models.json

models.json

files? We quarantined all 9 and one (Bragi) regenerated within 20 minutes. What triggers that, and can it be disabled?

Ollama and Kimi K2.5 cloud

Ask it these questions, in this order:

1. OpenClaw's Ollama provider is connecting to Ollama Cloud (ollama.com) using a thinking model (kimi-k2.5). The model returns content in the

thinking

thinking

field with an empty

response

response

field during streaming. The gateway hangs for 600 seconds then throws

FailoverError: LLM request timed out

FailoverError: LLM request timed out

. Is there a config setting to handle thinking model output, or does the Ollama provider only read the

response

response

field?
2. What is the correct

agents.list

agents.list

config to add fallback models to a per-agent primary override? Specifically: Agent Odin and Agent Mimir have

"model": {"primary": "openai-codex/gpt-5.4"}

"model": {"primary": "openai-codex/gpt-5.4"}

in their agents.list entries but no fallbacks. Does adding

"fallbacks": ["anthropic/claude-haiku-4-5"]

"fallbacks": ["anthropic/claude-haiku-4-5"]

models.json

models.json

files? We quarantined all 9 and one (Bragi) regenerated within 20 minutes. What triggers that, and can it be disabled?

Ollama and Kimi K2.5 cloud

Ollama and Kimi K2.5 cloud

Continue the conversation

Friends of the Crustacean 🦞🤝

Continue the conversation

Friends of the Crustacean 🦞🤝

Similar Threads