❓ FAQ & troubleshooting¶

Add the install dir to PATH and open a new shell:

• User install → ~/.local/bin
• System → $(brew --prefix)/bin or /usr/local/bin
• Windows → %LOCALAPPDATA%\slmcode\bin

hash -r
which slmcode
slmcode version

Re-run the installer. Corporate SSL inspection sometimes rewrites downloads. Try from a clean network or download the asset from GitHub Releases manually.

Formula SHAs track a specific release build. Update to latest main formula URL or reinstall after a formula sync commit.

1. Is the server up? (omlx start, ollama serve, LM Studio running…)
2. Does endpoint match reality (note /v1 vs bare host)?
3. Is the API key set for cloud providers?
4. Try:

curl -s "$ENDPOINT/models" | head
slmcode config
slmcode doctor

Per-agent endpoints register unique backend keys. Check .slmcode/agents/*.yaml and Studio → Agents. Global config can be fine while one specialist phones Mars.

Ollama uses its native API. Everything else is Chat Completions. Don't point provider: ollama at an OpenAI URL (or vice versa) unless you enjoy puzzles.

• Lower max_context_kb
• Pin atomic-coding
• Add a stern AGENTS.md
• Force a fresh explore once:

SLMCODE_FORCE_EXPLORE=1 slmcode run -v "…"

Expected on weaker SLMs. The harness repairs common breakage and retries. Raise think_passes / retries. Prefer models with decent tool-calling.

Check disk evidence in the Live feed. Prefer permission: review on important repos. Speculative cancel + disk rename acceptance exist specifically for flaky judges.

It shouldn't — explore reuse should kick in. If CONTEXT is empty or wiped, explore returns. Don't delete .slmcode/ for fun mid-project. The flywheel has feelings.

Rebuild the Go binary — assets are go:embed'd.

make build && ./bin/slmcode studio

Confirm .slmcode/queries/<id>/react/ exists. /resume needs a checkpoint from /stop or Ctrl+C — not a crashed machine mid-write with no flush.