hello-build-agent 0.5.6

Autonomous build assistant that generates Dockerfiles, ingests repos into Qdrant, and self-heals via LLM.

hello-build-agent

Autonomous DevOps side-kick
Clones any public repo, indexes docs to Qdrant, asks an LLM to craft the optimal Dockerfile, then builds, self-repairs and logs every step.

Changelog • License

---

✨ Key features

🧩	Description
LLM-driven Dockerfile	Uses OpenAI (gpt-4o-mini* by default) to propose a minimal, working Dockerfile based on real file-tree analysis.
Self-healing loop	If the image fails to build, the agent feeds the error back to the LLM → receives a patch → retries ― until success or user input.
Qdrant ingestion & search	README and file-tree chunks are embedded and stored in Qdrant; you can query them later (query_qdrant).
Quota guards	Two env-controlled limits: QUOTA_TOKENS, QUOTA_FAILS ― agent pauses and asks the human before overspending.
Live Rich logs	Timestamp · action · comment · token-counter, neatly aligned and colourised.
CI/CD ready	GitHub Actions → PyPI wheel + Docker Hub & GHCR images on each vX.Y.Z tag.

* Change the model with the MODEL env var or --model CLI flag.

---

🚀 Quick start

./quickstart.sh

build_logs/ and workspace/ are mounted on the host for inspection.

---

🛠️ CLI flags & env vars

CLI flag / env	Default	Purpose
--docker-host / DOCKER_HOST	tcp://dind:2375	Where to talk to Docker daemon
--qdrant-url / QDRANT_URL	http://qdrant:6333	Qdrant instance
--model / MODEL	gpt-4o-mini	OpenAI model name
QUOTA_TOKENS	10000	Ask human after this many tokens
QUOTA_FAILS	3	Ask human after N consecutive build errors
DISABLE_HITL	false	true → fully autonomous, no prompts

---

🧭 Architecture (90 sec overview)

CLI ─┬─ SystemPrompt (strict JSON contract)
     │
     ├─ loop.py  (LLM ↔ actions ↔ tracker/logs)
     │     ├─ tracker  ─── prints Rich log & enforces quotas
     │     ├─ actions  ─── clone, ingest, build, …
     │     └─ logger   ─── file logs (.log) for post-mortem
     │
     └─ docker_io.py  ── minimal, tail-aware build helper

---

📜 License

This project is distributed under the MIT License (see LICENSE).

---

Made with ❤️ & LLM-powered automation.