Runbook — set up Strix and credentials
Owner: whoever administers the pentesting toolchain · Frequency: once per
machine, revisit when rotating keys · Last updated: 2026-08-14 · Last run: verified against samples/strix/README.md and AGENTS.md, not executed against a live Docker daemon in this sandbox (no Docker available here).
Purpose
Get Strix installed, Docker verified, and an LLM provider credential wired up so a scan can actually run.
Prerequisites
- [ ] Docker installed and running.
- [ ] An LLM API key from a supported provider (OpenAI, Anthropic, Google, etc.) — or a ChatGPT Plus/Pro subscription if using the subscription auth flow instead.
- [ ] Outbound network access (to pull the sandbox image and reach the LLM provider).
Procedure
Step 1 — Install the CLI
curl -sSL https://strix.ai/install | bash
Expected result: strix is on PATH; strix --version (or equivalent) prints a
version string.
If it fails: check the installer's output for a permissions or network error; the
CLI can alternatively be run from source per CONTRIBUTING.md (uv run strix ...)
in ../../../samples/strix/.
Step 2 — Configure the LLM provider
export STRIX_LLM="openai/gpt-5.4" # any LiteLLM model id — see docs/llm-providers/ in the source repo
export LLM_API_KEY="your-api-key"
Expected result: both env vars are set for the shell session.
If it fails: double check the model id matches LiteLLM's naming convention for
your provider (anthropic/..., bedrock/..., vertex_ai/..., etc.) — see
strix/config/models.py (../../modules/config.md) for how this is resolved.
Optional overrides, if needed:
export LLM_API_BASE="your-api-base-url" # self-hosted models, e.g. Ollama, LM Studio
export PERPLEXITY_API_KEY="your-api-key" # enables the web_search tool
export STRIX_REASONING_EFFORT="high" # default: high; quick scans default to medium
Step 3 (alternative) — Sign in with a ChatGPT subscription instead of a metered key
strix auth login chatgpt
export STRIX_LLM="chatgpt/gpt-5.4"
strix auth status # confirm the active sign-in
Step 4 — Confirm settings persisted
Strix automatically writes resolved settings to ~/.strix/cli-config.json after a
successful run/preflight, so the env vars above don't need to be re-exported every
session (see ../../modules/config.md).
Verification
- [ ]
strix --target ./some-throwaway-dir --scan-mode quick --max-budget 1 -nstarts without an authentication or Docker error (it's fine if it stops quickly on budget — that confirms the plumbing works).
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| "Docker is not running" | Docker Desktop/daemon not started | Start Docker, retry. |
| Auth/connection error to the LLM provider on first run | Bad key, wrong model id, or no network | Re-check STRIX_LLM/LLM_API_KEY; warm_up_llm() (../../modules/interface.md) is what surfaces this. |
| Image pull is slow/fails | First run always pulls the sandbox Docker image | Retry with better network; check registry access. |
Rollback
Nothing persistent is created by setup beyond ~/.strix/cli-config.json and the
pulled Docker image — delete ~/.strix/cli-config.json to reset resolved config, or
docker rmi the sandbox image to force a re-pull.
Escalation
| Issue | Contact |
|---|---|
| LLM provider account/billing | Whoever owns the org's LLM provider account |
| Docker infra on shared machines | Platform/infra team |
History
| Date | Change |
|---|---|
| 2026-08-14 | Runbook authored against usestrix/strix commit 8ca0c4a9. |