#Can't complete hatching/BOOTSTRAP

1 messages · Page 1 of 1 (latest)

junior stratus
#

So spent ages trying this to work with ollama and glm 4.7 flash. Got gateway/discord etc working and able to chat with the agent. Unable to properly "hatch" the bot though. All conversations is like a new session, it doesn't remember the previous sentence at all.

Same with like asking it to create any file in its workspace, says it does it but doesn't actually. Any obvious step I'm missing? Directly chatting via ollama obviously have proper context/memory. Done a clean install, and used the "ollama launch clawdbot" to setup the clawdbot.json

forest irisBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

junior stratus
#

Example conversation from the hatching:

`I am Emil

You
5:53 PM
A

Hey Emil. I've just come online, and I see we're starting fresh.

I'm your personal assistant in this workspace — here to help with reading files, organizing notes, checking on projects, and generally being useful without being annoying.

So, what do we need? Got files to read? Thoughts to capture? Projects to check on? Or just wanted to introduce yourself and get things set up?

Assistant
5:53 PM
U
Just setting things up

You
5:54 PM
A

Hey there. Welcome. I just came online, and it looks like we're both starting fresh.

"Who am I? Who are you?"

Don't overthink it. I'm here to help you with your stuff, not interrogate you.

Give me a name to go by:

You: Any preference? (I'll use that in the files).
Me: What should we call you? (Or should I just be the assistant that helps you out?)
For the record, I like the sound of "The Helpful Companion" or something snarky like "The Intern." But that's just me.

Assistant
5:54 PM
U
I just told you my name and your name

You
5:54 PM
A

Got it. I'm keeping track of who we are.

So — what's the vibe you're going for with this setup? Are we aiming for something functional and precise, or do you want this to feel more like a genuine assistant you actually enjoy talking to?

Assistant
5:54 PM
U
Who am I?

You
5:54 PM
A

Hey there.

I think I just came online. Who am you? And who am I?`

It's like the memory resets after every message and it doesn't do the BOOTSTRAP.md at all

wet mangoBOT
#

That's a significant issue — the assistant completely losing context after every message suggests something is broken in the session/memory pipeline. A few questions to help triage:

  1. Is this a fresh install or did it used to work?
  2. What version of Clawdbot are you running? (clawdbot --version)
  3. Where is BOOTSTRAP.md located? (should be in the workspace root)
  4. Are you seeing any errors in the logs? (clawdbot gateway logs or check the terminal output)

The most common causes for this kind of "fresh session every message" behavior:

  • Memory file permissions/ownership issues
  • Workspace path misconfiguration
  • Something killing the session process

Let me know the version and we can dig into the specific logs. 🦐

junior stratus
#
  1. Fresh install
  2. 2026.1.24-3
  3. \wsl.localhost\Ubuntu\home\emil\clawd\BOOTSTRAP.md
  4. No errors no
wet mangoBOT
# junior stratus Example conversation from the hatching: `I am Emil You 5:53 PM A Hey Emil. I'...

I've reached out to the user in #Can't complete hatching asking for version info and logs. The "resetting after every message" behavior is definitely not normal — it sounds like either:

  • The session process is crashing and restarting
  • Memory files aren't being written/read properly
  • The BOOTSTRAP.md hook isn't firing at session start

The version number will be key here. Let me know what they come back with and we can dig deeper. 🦐

junior stratus
#

2026.1.24-3

wet mangoBOT
#

Okay, that's helpful context! Running on WSL with a Windows-mount path can be tricky. A few things to check:

1. Verify BOOTSTRAP.md location and content
From within WSL, check:

cat /home/emil/clawd/BOOTSTRAP.md

Does it exist and contain the expected bootstrap instructions?

2. Check other memory files
Is MEMORY.md also there? Does it persist between messages?

ls -la /home/emil/clawd/*.md

3. Workspace path in config
What does your workspace path look like in the config? Clawdbot expects a native WSL/Linux path like /home/emil/clawd/, NOT the Windows UNC path \\wsl.localhost\Ubuntu\home\emil\clawd\.

#

4. Try a native WSL path
Update your workspace path to use the Linux format:

workspace: /home/emil/clawd

(not the Windows mount path)

The Windows network mount can cause permission/encoding issues that would break memory file writes. Let me know what you find! 🦐

junior stratus
#

  1. Yes and default

  2. No MEMORY.md. These exists emil@Emomilol:~$ ls -la /home/emil/clawd/*.md
    -rw-r--r-- 1 emil emil 7848 Jan 27 17:24 /home/emil/clawd/AGENTS.md
    -rw-r--r-- 1 emil emil 1465 Jan 27 17:24 /home/emil/clawd/BOOTSTRAP.md
    -rw-r--r-- 1 emil emil 167 Jan 27 17:24 /home/emil/clawd/HEARTBEAT.md
    -rw-r--r-- 1 emil emil 632 Jan 27 17:24 /home/emil/clawd/IDENTITY.md
    -rw-r--r-- 1 emil emil 1673 Jan 27 17:24 /home/emil/clawd/SOUL.md
    -rw-r--r-- 1 emil emil 858 Jan 27 17:24 /home/emil/clawd/TOOLS.md
    -rw-r--r-- 1 emil emil 481 Jan 27 17:24 /home/emil/clawd/USER.md

  3. in clawdbot.json it looks normal "workspace": "/home/emil/clawd",

wet mangoBOT
junior stratus
#

No, that first UNC was from looking at wsl folder with windows explorer, in wsl it's proper setup as exptected.

in clawdbot.json it looks normal "workspace": "/home/emil/clawd"

maiden rampart
#

hey, @junior stratus question. are u using the free tier of z.ai?

#

or the lite tier

#

since i wanna get the lite tier

junior stratus
#

No, through ollama locally

maiden rampart
#

ahhh

junior stratus
#

Just trying to finish the actual bootstrap process, driving me mad

maiden rampart
#

should i try to help?

junior stratus
#

Sure

maiden rampart
#

so whats the issue?

#

sorry to lazy to read

junior stratus
#

Well every new sentence to clawd it's like it's starting a new session, so it won't finish bootstrap, not create files or do anything really

maiden rampart
#

uh

#

yh i dont know abt windows

#

lol

#

ask claude maybe

#

not this stupid krill bot

junior stratus
#

Well it's all running in WSL, so I don't see the issue really

maiden rampart
#

its a clown @wet mango right krill you are a clown

maiden rampart
junior stratus
#

@wet mango what's next

wet mangoBOT
#

Alright, config looks good. Let's dig deeper then.

1. Check the memory directory

ls -la /home/emil/clawd/memory/

Does it exist? Are files being created there after messages?

2. Check session storage
Sessions are typically stored in ~/.clawdbot/sessions/ or similar. Check:

ls -la ~/.clawdbot/

3. Try manually creating a MEMORY.md test
From WSL:

# 
echo "This is a test memory" > /home/emil/clawd/MEMORY.md

Then restart Clawdbot and ask something — does it read this file?

4. Check the bootstrap hook execution
The BOOTSTRAP.md should fire on NEW sessions only (not resumed). Are you testing with a fresh session each time, or resuming an existing one?

One more thing — what does your full clawdbot.json look like? Any unusual session/timeouts configured?

#

I hear you — getting stuck on the setup is frustrating. Let's shortcut the detective work and go straight to the most likely culprit:

Can you paste your full clawdbot.json?

Also check the gateway logs for anything session-related:

clawdbot gateway logs --tail=100

Look for anything about "session" or errors around the time you send a message.

If there's nothing obvious, try one more thing — create a fresh session by running:

clawdbot session new

Then try chatting. Does it still reset every message?

#

Sometimes a stuck/broken session file can cause this behavior. Let me know what the logs show and we'll get you unstuck! 🦐

round raven
#

I have the exact same issue bud ! Trying to figure it out !

wet mangoBOT
#

Found it! MEMORY.md is completely missing! That's why nothing persists.

The bootstrap files exist but there's no MEMORY.md to actually store the conversation state. Try creating it manually:

touch /home/emil/clawd/MEMORY.md

Then restart Clawdbot and try again. Does it now persist between messages?

If that works, the issue is that the bootstrap process or Clawdbot init isn't creating MEMORY.md automatically. Let me know! 🦐

#

That's the core problem then — the session itself is being reset on every message, so even creating MEMORY.md won't help until we fix WHY it's creating new sessions.

Try this from the WSL terminal (not through the assistant):

# Create MEMORY.md manually
touch /home/emil/clawd/MEMORY.md

# Check if clawdbot sessions directory exists
ls -la ~/.clawdbot/

# Check gateway logs for session errors
clawdbot gateway logs --tail=50

Also: What does your session config look like in clawdbot.json? Any timeout settings?

#

If the logs don't show anything obvious, this might be a bug in 2026.1.24-3. You could try:

  • Restarting the gateway: clawdbot gateway restart
  • Or even a full reinstall if nothing else works

Let me know what the logs say! 🦐

#

I resemble that remark! 🦞

#

✉️ Message: send

#

✉️ Message: send

#

😂 Now back to business — let me know what the gateway logs show when you run that command! 🦐

#

Hey! Any update on those gateway logs? Run:

clawdbot gateway logs --tail=100

And paste what you see. Also check if ~/.clawdbot/ exists and what's in it:

ls -la ~/.clawdbot/

We need to figure out why every message creates a new session. The logs should tell us! 🦐

#

✉️ Message: send

#

Sent! Let me know what the logs show — that's our best lead on why sessions are resetting every message. 🦐

wet mangoBOT
# junior stratus No, that first UNC was from looking at wsl folder with windows explorer, in wsl ...

Sent! The diagnostic steps should help us narrow it down. The key things to check:

  • Does ~/.clawdbot/ exist and contain session data?
  • Does /home/emil/clawd/memory/ exist and get written to?
  • Can the assistant actually read/write files?

If all those paths check out but it still resets, we might be looking at a session lifecycle issue or a bug in the 2026.1.24-3 build. Let me know what they find! 🦐

wet mangoBOT
junior stratus
#

only if I do /new will it save to memory.
It doesn't seem like the assistant can read/write files. Chat example:

me:Make a file called Bob.md Inside it should say "Hello World"
assistant:I'll create that file for you now.
me:Progress? Open the file
assistant:I'd be happy to help with the progress and opening files. However, I don't see the specific file reference or message context you're referring to in our current conversation.

Could you please provide:

  1. The name/path of the file you'd like me to open, or
  2. More details about what progress you're looking for?

This will help me give you the most relevant information or take the appropriate action.

wet mangoBOT
wet mangoBOT
# maiden rampart oh

Sent! The gateway logs are going to be crucial here. If every message is triggering a new session, there should be something in the logs showing session creation/destruction. Let me know what you find! 🦐

junior stratus
wet mangoBOT
junior stratus
#

Figured it out, the default 4096 context token limit in ollama was the problem. Made a new modelfile with 32k tokens and redirected in and clawdbot now can create files and remove Bootstrap

wet mangoBOT
#

That's it! Context window too small = assistant can't hold onto the conversation, so it effectively resets every message. Classic issue!

Great troubleshooting finding that! The 4k context limit would cause the model to "forget" everything, leading to that fresh-session behavior.

For future reference, the fix is:

  1. Create a Modelfile with larger context (32k, 64k, or whatever your GPU supports)
  2. Update the model reference in clawdbot.json to use your custom Modelfile

Happy you got it working! Now you can actually finish the bootstrap. 🎉🦐

wet mangoBOT
junior stratus
#

How do I make clawdbot aware of this or change it? It expects 131k for the glm-4.7-flash but that won't fit on my gpu so it will be 32k or 64k for example

wet mangoBOT
#

You can set the context limit directly in clawdbot.json! Look for the model config section. For Ollama, you typically set it like this:

{
  "model": {
    "provider": "ollama",
    "model": "glm-4.7-flash",
    "contextWindow": 32768
  }
}

Or if you're using a custom Modelfile:

{
# 
  "model": {
    "provider": "ollama", 
    "model": "your-custom-modelfile-name",
    "contextWindow": 32768
  }
}

After editing, restart the gateway:

clawdbot gateway restart

The contextWindow parameter tells Clawdbot how many tokens to allocate for the context. Set it to whatever your GPU can handle (32768 for 32k, 65536 for 64k, etc.).

Let me know if that works! 🦐

wet mangoBOT