FAQ & Glossary (AKIOS V1.0)

Quick FAQ

Is AKIOS an agent framework? No. It’s a security cage that runs agents/workflows with strict policies, sandboxing, redaction, and audit.
What’s the strongest setup? Pip on native Linux (cgroups v2 + seccomp-bpf). Docker wrapper is strong/policy-based across macOS/Win/Linux.
Where are my outputs? data/output/run_* per execution; audit logs in audit/audit_events.jsonl.
How do I switch to real APIs? --real-api or set AKIOS_MOCK_LLM=0 and ensure API key + network_access_allowed=true.
How do I check security status? akios status --security and akios audit verify.
What if a workflow fails validation? Ensure each step has agent, action, parameters; use allowed agents only; see Policy Schema.

Security cage: The enforced sandbox + policies + redaction + audit layer AKIOS wraps around every workflow.
Policy-based vs kernel-hard: Policy-based (Docker, macOS/Win) uses container isolation; kernel-hard (Linux pip) adds seccomp-bpf + cgroups v2 on host.
allowed_paths: Explicit filesystem allowlist for the filesystem agent; anything else is denied.
PII redaction: Real-time masking/hashing/removal of sensitive patterns on inputs/outputs.
Merkle audit: Tamper-evident log chain; akios audit verify checks integrity.
mock_llm / --real-api: Mock mode returns synthetic outputs; real API mode uses provider endpoints with budgets/redaction intact.
Budget/loop kill-switch: Hard limits on cost/tokens and loop detection to stop runaway workflows.
Agents: Four primitives—filesystem, http, llm, tool_executor—used to compose workflows; no arbitrary code execution outside them.