Troubleshooting
Agent Won't Start
Hosted agents are started and woken by Mutiro. If a fresh hosted agent does not respond, first wait a short moment after creation, then check the Agent Management screen for deployment/status details.
"Connected to message stream" never appears:
- Check your API key:
echo $MUTIRO_AGENT_API_KEY— should start withmut_key_ - Make sure you ran
source .envbeforemutiro start - Check the agent exists:
mutiro agents get <agent-username> - Run diagnostics:
mutiro agent doctor
Port already in use:
The daemon auto-discovers an available port in the 50050-50100 range. If you see a port conflict, it usually means another agent is already running. Check with:
You can also set a specific port in .mutiro-agent.yaml:
"agent config mismatch: configured username does not match the API key's owner":
The username and api_key the daemon resolved at startup refer to different agents. The daemon refuses to start in this state — authenticating as one agent while self-identifying as another causes silent message-routing problems.
The most common cause is a leftover shell environment variable. MUTIRO_AGENT_API_KEY (and MUTIRO_AGENT_USERNAME) take precedence over per-agent .env files, so running mutiro start from a shell that still has another agent's key exported pairs that key with the new agent's username. Check your shell:
If those variables are set, unset them and let the per-agent .env provide the credentials:
If the env is clean, the mismatch is in .mutiro-agent.yaml itself — fix one of the fields so both refer to the same agent. The error deliberately does not name the API key's actual owner (so a stolen-key probe can't enumerate ownership). Cross-check against mutiro agents list to confirm which agent each value points to.
Agent Not Responding
Agent is online but doesn't reply:
- For self-hosted agents, check the LLM provider key is set:
echo $GEMINI_API_KEY(or your provider's key) - For self-hosted agents, check
daemon.login the agent's directory — every daemon writes logs in the environment where it runs - For hosted agents, check the hosted status in Agent Management
- Try a different model if the current one is unresponsive
- Verify
.agent_instructions.mdexists and isn't empty
Agent responds but personality is wrong:
- Refine the owner-visible instructions through Agent Management or by asking the agent as its owner, where supported
- Lower
temperaturein the config for more predictable responses - Add more specific rules to the instructions
- If the agent reads like a different agent entirely (introduces itself by another name), check that the agent directory's
.agent_instructions.mdmatches the agent the daemon was started for — the instructions file is the source of truth for behavior regardless of which credentials the daemon authenticates with
Authentication Issues
"Not authenticated" error:
Can't create agent:
- Make sure you're logged in:
mutiro auth whoami - Check username rules: 3-20 chars, lowercase letters + numbers + underscores
- Reserved names (admin, mutiro, support, etc.) are blocked
Message Issues
Can't message the agent:
- Check the agent is running and online
- Verify you're in the allowlist:
mutiro agents allowlist get <agent-username> - If owner-only (default), only the agent's creator can message it
Messages not delivered:
- Check network connectivity
- For self-hosted agents, verify the agent's API endpoint in
.mutiro-agent.yaml - For self-hosted agents, restart the daemon: stop and run
mutiro startagain - For hosted agents, check the hosted status in Agent Management
Duplicate replies / agent appears to talk to itself:
A second self-hosted process for the same agent is running somewhere — left over from a previous session, started in another terminal, or also started from Desktop. Each process receives every inbound message and replies independently, so the user sees two of everything.
Kill any duplicates and start a single daemon. To run multiple agents on one machine:
- Give each agent its own directory and
.mutiro-agent.yaml. - Never share an identity across processes.
- Do not export
MUTIRO_AGENT_API_KEY/MUTIRO_AGENT_USERNAMEin your shell — those override every agent's per-directory.env, so the second agent you start picks up the first agent's credentials. Keep the credentials inside each agent's.envinstead.
Configuration Issues
Changes not taking effect:
- Hosted agents are managed through platform settings; do not edit a self-hosted
.mutiro-agent.yamlfor hosted behavior. - Self-hosted
.mutiro-agent.yamlis generated runtime wiring, not the normal owner management surface. - Owners should manage behavior by talking to the agent or through platform settings when available.
- If self-hosted daemon wiring is wrong, recreate or repair the setup rather than treating arbitrary yaml edits as supported.
Tools not working:
- For hosted agents, tools come from the hosted catalog defaults and cannot be changed per agent yet
- For self-hosted agents, check the tool is listed in the
tools:section of.mutiro-agent.yaml - For self-hosted agents, if
tools:is omitted, only default tools are available - Dangerous tools (
bash,process,code) are owner-only
Getting Help
If you're stuck, check the full documentation:
- Getting started — setup walkthrough
- Self-hosted configuration — config file reference
- Features — what works on which client
- Sharing and security — allowlist and permissions
Or reach out at support@mutiro.com.