v1.4.0

[1.4.0] — 2026-06-04

Added

• AWS Bedrock provider — run Crash Handler and QA Agent via anthropic.AnthropicBedrock
  using an IAM role; no API key stored. Cross-region inference profiles are resolved
  automatically from the AWS region prefix (us., eu., ap.).
• OpenRouter provider — OpenAI-compatible API backend; set OPENROUTER_API_KEY and
  provider: openrouter per agent to use any model on OpenRouter's catalogue.
• OpenCode and Goose CLI backends — Dev Agent can now dispatch to the OpenCode or
  Goose CLI as alternatives to Claude Code.
• Per-agent LLM configuration — config.yaml now has an agents: block (replacing
  the global llm: block). Each agent has its own provider and model. Override at
  runtime with HELIX_<AGENT>_PROVIDER and HELIX_<AGENT>_MODEL.
• core/preflight.py — startup validation that checks at least one LLM provider key
  is set (or Bedrock is configured via IAM). Warns on missing optional env vars.
• LangSmith tracing (opt-in) — when LANGSMITH_API_KEY and LANGSMITH_TRACING=true
  are set, every complete() call is traced with provider, model, and token usage.
• openai>=1.55 dependency for the OpenRouter and Ollama backends (replaces the previous
  httpx-based Ollama implementation).

Changed

• core/llm.py — unified router now dispatches to seven backends: anthropic, bedrock,
  openrouter, ollama, claude-code, opencode, goose. Anthropic backend retries
  with claude-haiku-4-5-20251001 on 529 Overloaded responses.
• core/config.py — LLMConfig / get_llm_config() replaced by AgentConfig /
  get_agent_config(agent). Added get_bedrock_region().
• agents/dev/agent.py — complete_tdd(cwd=...) replaced by complete(agent="dev", cwd=...);
  provider is now determined by the dev agent's config entry.
• config.yaml — llm: block replaced by agents: block; settings.aws_bedrock_region
  added.

v1.3.0

Added

• Railway deployment — one-click deploy button in README; railway.toml,
  entrypoint.sh, and railway-deploy.sh added so the full stack can be
  deployed to Railway without local Docker Compose.
• entrypoint.sh supports three modes: docker-compose command: args,
  START_COMMAND env var (per-service on Railway), or all-agents-in-one-container
  fallback for single-service deployments.

v1.2.0

Security

• Upgraded langsmith to fix a token vulnerability in streams.

v1.1.1

Security fixes

• Information exposure — raw exception messages are no longer included in Slack responses; errors are logged server-side only (#1)
• Workflow permissions — CI workflow now declares permissions: contents: read to enforce least-privilege on the auto-injected GITHUB_TOKEN (#2)

v1.1.0

What's new

• Ollama / local LLM support — run Helix against any locally hosted model (Llama, Mistral, CodeLlama, etc.) via Ollama's OpenAI-compatible API. Set HELIX_PROVIDER=ollama and HELIX_MODEL=llama3.2 to switch providers without touching agent code.
• Contact & Cloud CTA — added contact section and Helix Cloud sales link to the landing page.
• Updated .env.example, config.yaml, README.md, and core/llm.py / core/config.py accordingly.

Helix v1.0.0

Helix v1.0.0 — Initial release

Autonomous incident response pipeline: Sentry/Rollbar crash to fix suggestion, with human review via Slack.

Pipeline

1. Crash Handler — ingests Sentry/Rollbar webhooks, classifies crash severity and language
2. QA Agent — generates a failing test case, opens a GitHub Issue
3. Dev Agent — fetches source files, posts an LLM-generated fix suggestion to the Issue
4. Notifier — sends a Slack notification with crash context and Issue link

What's included

• Sentry + Rollbar webhook ingestion (HMAC-verified)
• Multi-language test generation (Python, JS/TS, Ruby, Java/Kotlin, Go)
• Fix suggestions via Anthropic API, single shared model (claude-sonnet-4-6)
• Redis Pub/Sub event bus
• Slack notifications
• Self-hosted via Docker Compose

Quick start

git clone https://github.com/88hours/helix-community.git
cd helix-community
cp .env.example .env
docker compose up --build