README

Open-source Laravel package that detects recent application errors, asks an LLM for a fix proposal (OpenAI, Anthropic, Hugging Face, or Ollama), validates safety rules, applies fixes in hybrid mode, and sends healing reports.

What It Does

Detects recent app errors and builds a repair context.
Requests a fix proposal from a configured LLM provider.
Validates proposed changes against strict safety rules.
Applies validated fixes in dry-run or auto-apply mode.
Sends reports to chat/monitoring channels.

Quick Start

1) Install

composer require self-heallm/laravel-self-heallm
php artisan vendor:publish --tag=self-heal-config

2) Configure

Main config file: config/self-heal.php

Supported sections:

Provider selector: llm.provider (openai, anthropic, huggingface, ollama)
OpenAI: openai.base_url, openai.token, openai.model, openai.timeout
Anthropic: anthropic.base_url, anthropic.token, anthropic.model, anthropic.timeout, anthropic.version
Hugging Face (OpenAI-compatible endpoint): huggingface.base_url, huggingface.token, huggingface.model, huggingface.timeout
Ollama (OpenAI-compatible endpoint): ollama.base_url, ollama.token, ollama.model, ollama.timeout
Telegram: telegram.bot_token, telegram.user_id, telegram.enabled
Slack: slack.enabled, slack.webhook_url
Generic webhook: webhook.enabled, webhook.url, webhook.token
Sentry: sentry.enabled, sentry.dsn, sentry.environment
Modes: enabled, auto_apply, dry_run
Safety: allowed_paths, forbidden_paths, max_files_per_fix
Deduplication: deduplication.enabled, deduplication.ttl_seconds, deduplication.store_path
Model context size: context.max_file_chars
Reporting: JSONL fallback path via reporting.json_path

Example .env values:

SELF_HEAL_ENABLED=true
SELF_HEAL_AUTO_APPLY=false
SELF_HEAL_DRY_RUN=true
SELF_HEAL_LLM_PROVIDER=openai
SELF_HEAL_OPENAI_BASE_URL=https://api.openai.com/v1
SELF_HEAL_OPENAI_TOKEN=sk-...
SELF_HEAL_OPENAI_MODEL=gpt-4.1-mini
SELF_HEAL_ANTHROPIC_BASE_URL=https://api.anthropic.com/v1
SELF_HEAL_ANTHROPIC_TOKEN=
SELF_HEAL_ANTHROPIC_MODEL=claude-3-5-sonnet-latest
SELF_HEAL_HUGGINGFACE_BASE_URL=https://router.huggingface.co/v1
SELF_HEAL_HUGGINGFACE_TOKEN=
SELF_HEAL_HUGGINGFACE_MODEL=openai/gpt-oss-120b
SELF_HEAL_OLLAMA_BASE_URL=http://localhost:11434/v1
SELF_HEAL_OLLAMA_TOKEN=
SELF_HEAL_OLLAMA_MODEL=llama3.1
SELF_HEAL_TELEGRAM_BOT_TOKEN=123:abc
SELF_HEAL_TELEGRAM_USER_ID=123456789
SELF_HEAL_SLACK_ENABLED=false
SELF_HEAL_SLACK_WEBHOOK_URL=
SELF_HEAL_WEBHOOK_ENABLED=false
SELF_HEAL_WEBHOOK_URL=
SELF_HEAL_WEBHOOK_TOKEN=
SELF_HEAL_SENTRY_ENABLED=false
SELF_HEAL_SENTRY_DSN=
SELF_HEAL_DEDUP_ENABLED=true
SELF_HEAL_DEDUP_TTL_SECONDS=600
SELF_HEAL_CONTEXT_MAX_FILE_CHARS=12000

Provider notes:

openai: default provider, no migration needed for existing setups.
anthropic: uses native Anthropic Messages API.
huggingface: in this version, use an OpenAI-compatible Hugging Face endpoint.
ollama: in this version, use Ollama's OpenAI-compatible endpoint (/v1/chat/completions).

3) Run

php artisan self-heal:run

Security Model

Accepts fixes only for whitelisted paths.
Rejects forbidden paths even if suggested by the model.
Rejects empty replacement targets.
Applies validated patches automatically only with dry_run=false and auto_apply=true.
Skips duplicate error fingerprints during deduplication TTL.

Reporting Channels

Telegram notifications.
Slack incoming webhook notifications.
Generic webhook (POST) notifications.
Sentry envelope event reporting.
JSON fallback report when enabled.

self-heallm / laravel-self-heallm

Maintainers

Package info

Statistics

Security