Security, model routing, overnight maintenance, knowledge graphs, and self-improvement. Five shell-native tools. One install. No cloud lock-in.
Each solves a real problem standalone. Together they create compound capabilities none has alone.
Pre-execution command interception. Blocks destructive ops, secrets leaks, PII exposure at the shell level before they run.
Cost-optimized LLM selection. 6 tiers from local Ollama to cloud APIs. One eval sets $MODEL, $PROVIDER, $API_KEY.
Autonomous overnight maintenance. Scans repos, patches lint errors, fixes vulnerabilities, opens PRs while you sleep.
Codebase to knowledge graph. Maps files, functions, imports as queryable JSON. Measure blast radius before you commit.
Self-improving agent framework. Tracks what your agents do, learns from failures, improves behavior over time.
Individual tools solve individual problems. The stack creates compound capabilities.
Solo: GuardRail blocks it.
Stack: GuardRail blocks it. Autonomie OS learns the pattern. Night Shift adds a regression test.
Solo: Graphify maps it.
Stack: Graphify maps it. Model Router picks the right LLM for review. GuardRail validates auth middleware.
Solo: Night Shift patches it.
Stack: Night Shift patches it. GuardRail validates the fix. Graphify updates the knowledge graph.
Solo: Model Router picks one.
Stack: Router picks based on Graphify's complexity score. Autonomie OS remembers which models worked best.
Pre-execution security. Shell-native. Local-first.
| Capability | AgentStack | LiteLLM | Guardrails AI | DIY |
|---|---|---|---|---|
| Pre-execution security | Yes | No | Prompt only | Manual |
| Model routing | 6 tiers | Yes | No | Manual |
| Overnight automation | Yes | No | No | Crons |
| Knowledge graphs | Yes | No | No | Separate |
| Self-improvement | Yes | No | No | No |
| Local-first (Ollama) | Yes | Partial | No | Depends |
| Zero dependencies | bash+jq | Python | Python | Varies |
| Test suite | 345 tests | Yes | Yes | Yours |
Requires bash 4+, git, curl, jq. Linux and macOS.
Or: docker compose up
Every component is MIT licensed. Pro tiers add support and enterprise features.
For solo developers and small teams.
For teams shipping AI agents to production.
For organizations with compliance needs.