FAQ

Is my data safe with the AGPL license?

AGPL-3.0 covers the Reelier engine itself (the parser, runner, and CLI) — not your data. Your traces, your SKILL.md files, and your run records are yours; the license has no claim on them. See the Licensing section of the README.

What data does Reelier collect?

The CLI itself collects nothing — no telemetry, no phone-home. Recording writes trace files to your own .reelier/ directory; nothing leaves your machine unless you run reelier push. Full detail in the privacy policy.

What can Reelier replay, and what can’t it?

Can: a deterministic sequence of tool calls — the exact same MCP tool, same filled arguments, checked against a fresh assertion every time. That covers most agent busywork: lookups, status checks, data pulls, idempotent writes.

Can’t:open-ended generation or judgment calls — anything where the “right” output isn’t a checkable assertion but a model’s free-form reasoning. Reelier records tool calls, not thought — a step is either a concrete, assertable tool call or it’s not something Level 0 can express. Escalation (L1/L2) uses an LLM only to patch one broken assertion or re-derive one failed step’s args, not to re-reason through the whole task.

Is it brittle like RPA?

No — it replays tool calls, not pixels or DOM selectors, so there’s nothing to break on a UI redesign. Every step carries its own assertion, so a broken step fails loudly instead of silently passing (a step with zero assertions is honestly reported as unchecked, never passed). And when the world genuinely drifts underneath a skill, the self-heal ladder (L1/L2) patches it once and writes the fix back — the same drift never has to escalate twice.

What’s the determinism guarantee, exactly?

Level 0 replay is deterministic for the recorded tool-call sequence: same tool, same filled args, evaluated against a fresh result each run. It is not a guarantee that the downstream service itself returns the same data forever — an assertion can still fail if the world changed. What’s guaranteed is that Reelier never silently reports a check it didn’t actually run, and never fabricates a passing result.

Is my LLM API key safe?

Yes — it’s a local CLI flag or environment variable (--llm-api-key / REELIER_LLM_API_KEY), sent directly to whichever endpoint you configured with --llm-base-url. It is never received or stored by Reelier Cloud, and it’s only ever used when a step actually escalates — a run whose skill never diverges never touches it.

Where can I read more?

How it works covers the full record/compile/replay/heal loop. SPEC.md is the normative, RFC-style reference for every file format. See also the privacy policy and terms of service.