kimi-cli接入本地ollama报错:✨ hi LLM provider error: The API returned an empty response.

[minghe-zty]

What version of Kimi Code CLI is running?

1.17.0

Which open platform/subscription were you using?

/login

Which model were you using?

ollama

What platform is your computer?

Darwin 25.1.0 arm64 arm

What issue are you seeing?

我想替换kimi-cli的api为自部署kimi 2.5模型api,尝试后发现会报错 LLM provider error: The API returned an empty response.;替换api为第三方中转openai api也会报错 LLM provider error: The API returned an empty response.,尝试替换api为本地ollama,使用ollama中kimi-k2.5:cloud 可以正常运行 切换为ollama中llama3.2,会报错LLM provider error: The API returned an empty response.

以下是我的config.toml:
default_model = "llama3.2"

[models."llama3.2"]
provider = "ollama"
model = "llama3.2"
max_context_size = 32768
capabilities = ["image_in"]

[models."kimi-k2.5:cloud"]
provider = "ollama"
model = "kimi-k2.5:cloud"
max_context_size = 32768
capabilities = ["image_in"]

[providers.ollama]
type = "openai_legacy"
base_url = "http://localhost:11434/v1"
api_key = "ollama"

以下是输出:
~ kimi
╭───────────────────────────────────────────────────────╮
│ │
│ ▐█▛█▛█▌ Welcome to Kimi Code CLI! │
│ ▐█████▌ Send /help for help information. │
│ │
│ Directory: ~ │
│ Session: 618fc273-210d-43bd-8441-d62952e30bf7 │
│ Model: llama3.2 │
│ Tip: send /login to use our latest kimi-k2.5 model │
│ │
│ Tip: Kimi Code Web UI, a GUI version of Kimi Code, │
│ is now in technical preview. │
│ Type /web to switch, or next time run kimi we │ │ b directly. │
│ │
╰───────────────────────────────────────────────────────╯
✨ hi
LLM provider error: The API returned an empty response.

✨ /model
Select a model (↑↓ navigate, Enter select, Ctrl+C cance
l):

1. kimi-k2.5:cloud (ollama)
   2. llama3.2 (ollama) (current)
   Switched to kimi-k2.5:cloud with thinking off. Reloading.
   ..
   ╭───────────────────────────────────────────────────────╮
   │ │
   │ ▐█▛█▛█▌ Welcome to Kimi Code CLI! │
   │ ▐█████▌ Send /help for help information. │
   │ │
   │ Directory: ~ │
   │ Session: 618fc273-210d-43bd-8441-d62952e30bf7 │
   │ Model: kimi-k2.5:cloud │
   │ Tip: send /login to use our latest kimi-k2.5 model │
   │ │
   │ Tip: Kimi Code Web UI, a GUI version of Kimi Code, │
   │ is now in technical preview. │
   │ Type /web to switch, or next time run kimi we │ │ b directly. │
   │ │
   ╰───────────────────────────────────────────────────────╯
   ✨ hi，what's your name
   • Used Shell (sysctl -n hw.ncpu)
   • Used ReadFile (dijkstra.py)
   • 让我先查找一下相关的文件。
   • Used Glob (111)
   • Used Glob (111/.py)
   • I don't have access to previous conversation history. L
   et me look for any game repositories in the current wor
   king directory that you might be referring to.
   • Used Glob (game/**/)
   • Used Glob (new_game//)
   • Used Glob (anthropic_test//)
   • I'll check the existing dijkstra.py file first to see w
   hat's already there.
   • Used ReadFile (dijkstra.py)
   • I'll create a stunning, animated personal portfolio web
   site for you. Let me start by exploring the current dir
   ectory and then build this step by step.
   • Used Glob ()
   • Used ReadMediaFile (.kimi/avatar.jpg)
   File not found
   • I'll help you design and implement a simple 2D platform
   er game. Let me first explore the working directory to
   see what's already there, and then I'll create the game
   .
   • Used Shell (ls -la /Users/zty/game/ &...s empty or does
   n't exist")
   ⠦ Composing...
   context: 20.3% (6.7k/32.8k)

What steps can reproduce the bug?

本地安装ollama
运行ollama pull llama3.2
修改kimi-cli的config.toml:
default_model = "llama3.2"

[models."llama3.2"]
provider = "ollama"
model = "llama3.2"
max_context_size = 32768
capabilities = ["image_in"]

[models."kimi-k2.5:cloud"]
provider = "ollama"
model = "kimi-k2.5:cloud"
max_context_size = 32768
capabilities = ["image_in"]

[providers.ollama]
type = "openai_legacy"
base_url = "http://localhost:11434/v1"
api_key = "ollama"

再次启动kimi cli即可

What is the expected behavior?

期望kimi-cli可以接入其他api,至少可以接入自己部署的kimi k2.5

Additional information

No response

[minghe-zty]

昨天验证后发现

default_model = "MiniMax-M2.5"
default_thinking = false
default_yolo = false
default_editor = ""

[models."MiniMax-M2.5"]
provider = "minimax"
model = "MiniMax-M2.5"
max_context_size = 32768
capabilities = ["image_in"]

[providers.minimax]
type = "openai_legacy"
base_url = "https://api.minimaxi.com/v1"
api_key = "xxx"

[loop_control]
reserved_context_size = 4096
max_steps_per_turn = 30
max_retries_per_step = 2
max_ralph_iterations = 0

[services]

[mcp.client]
tool_call_timeout_ms = 60000

在简单场景下(hello world)可以运行,不知道使用高级功能会不会出问题

kimi -p "hi" --print --model llama3.2 --config-file ~/.kimi/config.toml

hi
TurnBegin(user_input='hi')
StepBegin(n=1)
CompactionBegin()
CompactionEnd()
StepInterrupted()
Error code: 502

估测在上下文压缩阶段出现问题

[lucky-lbc]

老哥，我怎么感觉最新版，还是没修复，是需要配置什么吗？ 程序直接崩溃

---

Brother, I feel like the latest version is still not fixed. Is there anything I need to configure? The program crashes directly

[minghe-zty]

crashes

你是没有办法接入本地ollama吗

---

crashes

Is there no way for you to access local ollama

[minghe-zty]

老哥，我怎么感觉最新版，还是没修复，是需要配置什么吗？ 程序直接崩溃

Brother, I feel like the latest version is still not fixed. Is there anything I need to configure? The program crashes directly

这是我的配置信息:

# ~/.kimi/config.toml
default_model = "kimi-k2.5-local"
default_thinking = true
default_yolo = false
default_editor = ""

[providers.ollama]
type = "openai_legacy"
base_url = "http://127.0.0.1:11434/v1"
api_key = "ollama"

[models."kimi-k2.5-local"]
provider = "ollama"
model = "kimi-k2.5:cloud"
max_context_size = 32768

# 可选：本地纯模型备用
# [models."llama3.2-local"]
# provider = "ollama"
# model = "llama3.2"
# max_context_size = 32768

[loop_control]
max_steps_per_turn = 30
max_retries_per_step = 2
max_ralph_iterations = 0
reserved_context_size = 4096
compaction_trigger_ratio = 0.85

[mcp.client]
tool_call_timeout_ms = 60000

再加这两行环境变量

export NO_PROXY=127.0.0.1,localhost
export no_proxy=127.0.0.1,localhost

验证命令：

kimi -p "hi" --print --model kimi-k2.5-local --config-file ~/.kimi/config.toml

我的结果:

hi
TurnBegin(user_input='hi')
StepBegin(n=1)
TextPart(type='text', text='Hello! How can I help you today?')
StatusUpdate(
    context_usage=0.290435791015625,
    context_tokens=9517,
    max_context_tokens=32768,
    token_usage=TokenUsage(
        input_other=9517,
        output=33,
        input_cache_read=0,
        input_cache_creation=0
    ),
    message_id='chatcmpl-550',
    plan_mode=False,
    mcp_status=None
)
TurnEnd()

tui测试: