Skip to content

v0.0.124: Frequent connection interruptions, 502 Bad Gateway, long latency and session stuck in Connecting #880

Description

@pacosoeda2

Environment
Freebuff CLI version: 0.0.124
OS: Windows 10 x64
Stable home fiber connection
Full mode
Models tested:
Kimi K2.7 Code
MiniMax M3
Summary

Starting today (yesterday everything worked normally), Freebuff has become significantly slower and is frequently losing connection during agent execution.

The issue affects multiple models and is reproducible even in a brand new empty directory.

Symptoms observed

  1. Large delay before responses start

Even a simple prompt such as:

Hello

takes approximately 30–60 seconds before the model actually starts generating.

The CLI itself reacts quickly:

Press Enter → "Thinking": around 3 seconds
"Thinking" → first generated output: 30–60 seconds
2. Happens with multiple models

I reproduced the same behaviour with:

Kimi K2.7 Code
MiniMax M3

Therefore it does not appear to be model-specific.

  1. Happens in an empty directory

I created a completely new empty folder and started a new Freebuff session.

Even there, a simple "Hello" took around 36 seconds before producing any output.

  1. Connection state changes during streaming

While the model is already producing output, the status sometimes changes like this:

Working

Connecting

Connected

Streaming continues

Sometimes it recovers.

Sometimes it never does.

  1. Internal agent failure

During execution I received the following error:

Agent run error:
Failed after 4 attempts.
Last error: Bad Gateway

The failing internal agent was:

basher

At the same time, another internal agent (code-reviewer-kimi) completed successfully.

  1. Automatic retry

After the failure, the agent automatically retried and displayed a message explaining that the agents had failed because of connection problems and that it was going to retry.

The second basher execution completed successfully.

However...

  1. Session remains stuck in Connecting

Even after:

basher completed

the global session status still remains:

Connecting...

for a long time.

Sometimes it later returns to:

Working...

without producing any new output.

Other times it never recovers.

  1. Timeout while switching models

At one point, switching from Kimi to MiniMax resulted in a timeout.

After restarting the CLI, it reported that another session was still active and required a takeover before I could continue.

Expected behaviour
Responses should start within a few seconds.
No unexpected transitions from Working to Connecting while streaming.
Successful agent completion should end the task normally.
No 502 Bad Gateway errors from internal agents.
Troubleshooting already performed
Tested with different models (Kimi K2.7 Code and MiniMax M3).
Tested inside a very large project (over 800 messages in the conversation).
Tested in a completely new empty directory.
Restarted the CLI.
Performed a session takeover after a timeout.
The problem persists.
Additional observations

This behaviour started today.

Yesterday the exact same workflow was working normally.

Because:

multiple models are affected,
it also happens outside any project,
internal agents report 502 Bad Gateway,
the connection is interrupted and automatically retried,
the session sometimes remains indefinitely in "Connecting",

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions