Skip to content

Barx 1.0.0 — Runtime Intelligence Layer

Latest
Latest

Choose a tag to compare

View all tags
@TheBarmaEffect TheBarmaEffect released this 13 Jun 09:29
· 3 commits to main since this release
v1.0.0
ca6047c

Barx 1.0.0 is the first stable release of the Barx Runtime Intelligence Layer.

Everything your code did, explained through evidence.

Barx watches a Python run, records structured evidence to an append-only log, and renders it as a report, a local workspace, and a release verdict — GREEN / AMBER / RED. AI writes code faster than teams can review it; Barx makes what a run actually did inspectable, so you can decide — with evidence — whether it's ready to ship.

The retired 2025 0.1.0 PyPI package ("Fast, CPU-only AI framework") was a different product. This is the rebuilt Runtime Intelligence Layer.

Install

pip install barx            # core
pip install "barx[api]"     # + API testing (httpx), optional
barx version                # -> barx 1.0.0

What's in 1.0.0

  • Trace — execution spans, timing, exceptions, recorded.
  • Verify — behavioral verification + a 20-rule AST risk scan (no code execution for the scan). Not formal verification.
  • API — ordered suites with status/latency/JSON-path/schema-lite assertions, token chaining, redaction. Not a Postman replacement.
  • Policy + Guard — observe / warn / strict over documented seams (eval/exec, subprocess, file delete, network…). Runtime guardrails, not a full sandbox.
  • Graph — project (AST), runtime, and failure graphs; JSON + Mermaid.
  • Drift + Replay — compare two runs (comparative evidence, not causal proof); replay is dry-run and GET-only by default.
  • Score + ReleaseGate — evidence-backed verdict with stated formulas and the blockers attached.
  • AI Runtime — LLMTrace (prompts/responses hashed by default), PromptGuard, Cost (an estimate from your price table, not exact billing).
  • AgentAudit — what an AI coding agent changed (files, deps, commands, tests) — file contents never stored.
  • Evidence Testing — Mock / Contract / AutoTest generated from recorded evidence (review required).
  • Studio — a local-only visual workspace (binds 127.0.0.1). The report is the source of truth; Studio is the viewer.
  • Integrations — GitHub Action (release gate + PR comment) and a VS Code MVP. CLI shells out only; no second evidence engine.

Where it fits in your workflow

See the new guide: docs/using-barx.md — inner loop, pre-PR, CI gate, editor, AI agents, API work, debugging, and a command-to-moment cheat sheet.

Honest boundaries

Not formal verification · not a full sandbox · not exact billing · not a cloud platform. Drift is comparative evidence, not causal proof; the failure graph is event-supported, not a root-cause claim. See docs/AUDIT.md and docs/CLAIMS.md.

Privacy

Local-first, no telemetry, no network calls unless you explicitly wire up the PR comment. Prompts/responses SHA-256 hashed by default; secrets redacted in every stored artifact; AgentAudit stores hashes/sizes/mtimes, never contents — all grep-tested in the suite.

Quality

884 tests, 91% coverage, ruff + format gates, CI across Python 3.10–3.12, and a self-checking release gate. Published to PyPI via Trusted Publishing (OIDC) with signed attestations — no tokens, no stored secrets.

Built by Karthik Barma · Aura.

Assets 2
Loading