Getting started with Helix
How to write a great prompt, choose autonomy, and review actions.
Help Center
How can we help?
Browse the Helix knowledge base for an instant answer, or open a ticket and we'll route it to a human within your plan's response window.
Open a ticket
Email support@launchverse.app with your team name, a clear description, and any relevant thread or run ID. Reply window varies by plan — see SLA below.
Check system status
Live status of the Helix brain, dashboard, sandboxes, and connected cloud providers. Subscribe to incidents from the status page.
FAQ
Answers to the most common pre-sales questions: pricing, quotas, supported frameworks, security, and cloud providers.
Knowledge base
Tap a category, then jump to the article below.
Repos & code
Connect GitHub, choose branches, review diffs, and publish PRs.
Sandboxes & compute
Run commands, install dependencies, and debug inside an isolated sandbox.
Cloud integrations
Connect MCP providers so Helix can deploy, debug, and manage infrastructure.
Security & trust
Secrets, permissions, approvals, and how Helix stays inside guardrails.
Billing & usage
Helix queries, sandbox compute, and provider API call quotas.
Reliability & limits
Context budgets, condensing, recovery, and avoiding runaway runs.
Reference
Docs, status, and the Helix changelog.
Articles
What is Helix?
+
Helix is an autonomous AI software engineer living in your team. Connect your repositories and cloud providers, then delegate coding, debugging, migrations, testing, PR creation, and deployments in plain English. Helix runs inside an isolated sandbox, reasons step by step, and only takes write actions at the autonomy level you choose.
How to write a good Helix prompt
+
Be specific about the goal, the repo or branch, and the acceptance criteria. For example: “Fix the failing login test in backend/auth.test.js, open a PR against main, and run lint and typecheck before publishing.” Helix will break this into read, plan, execute, verify, and publish steps.
Autonomy levels explained
+
Level 0 is read-only advice. Level 1 suggests code but does not edit. Level 2 edits files and runs tests. Level 3 can open PRs. Level 4 can also deploy and manage resources through connected MCP providers. You can switch level at any time in the Helix panel.
Approving and rejecting Helix actions
+
At lower autonomy levels Helix pauses before writes and shows a PROPOSED card. Approve to let it continue, Reject to cancel that step, or Amend to edit the command before it runs. At level 4 some actions like test execution happen inline, but destructive actions still ask.
Helix memory and context
+
Helix keeps short-term memory within a thread and learns your project conventions across sessions. Long conversations are condensed so Helix stays inside model context limits, with the original system prompt and recent turns always preserved.
Linking a GitHub repo
+
Install the Helix GitHub App on the repo from Dashboard → Integrations, then mention it in a prompt. Helix clones it into a sandbox on its first use.
Which branch should Helix use?
+
Helix defaults to the repo's default branch. You can override by saying “work on the vulnerable branch” or by passing a branch name in the prompt. When publishing fixes, Helix will create a new branch and open a PR against the branch you specify.
How Helix opens and updates pull requests
+
When a task is complete and checks pass, Helix pushes the sandbox state to a branch and opens a pull request with a summary of the changes. It never merges unless you explicitly ask it to.
Pre-PR checks: test, lint, typecheck, build
+
Helix detects your framework from package manifests and config files. It runs test, lint, typecheck, and build before publishing. If a script is missing but the config exists, Helix adds the missing script so the check can run. Repos without any matching tooling skip that check.
Helix couldn't push to my repo
+
Make sure the Helix GitHub App is installed and has write access. If the branch is protected, Helix will create a PR from a new branch instead of pushing directly. Check the tool result for the exact error.
What can Helix run in a sandbox?
+
Almost any command you would run locally: package installs, tests, builds, lint, git, and language-specific tools. The sandbox is an isolated container with internet access disabled by default for security.
Sandbox quotas and timeouts
+
Each plan has a daily sandbox compute quota and a per-command timeout. Free gets 30 minutes of compute per day; Pro, Team, and Enterprise get progressively more. Check Dashboard → Usage for live numbers.
Reusing a sandbox across turns
+
A sandbox stays alive for the duration of a thread and is reused across turns. If something breaks, you can ask Helix to create a fresh sandbox from the same repo.
Installing missing tools
+
If a tool is missing, Helix can install it with the package manager available in the sandbox (apt, npm, pip, cargo, etc.). For uncommon binaries it may ask you to confirm first.
Setting up the Coolify MCP provider
+
Go to Dashboard → Settings → MCP, add your Coolify API token and base URL, and enable the Coolify provider. Helix can then list projects, create applications, trigger builds, and fetch deployment status.
Setting up the AWS MCP provider
+
AWS and other MCP providers are configured the same way: add the API credentials in the MCP settings panel, enable the provider, and Helix will expose the provider's tools when relevant to your prompt.
Helix deployed to the wrong app
+
Make sure the project name is explicit in your prompt, e.g. “deploy to the ticketing-apiCoolify app.” Helix will list matching apps and ask for confirmation if it finds more than one.
Deploy quotas and limits
+
Deploy actions consume provider API calls and sometimes sandbox compute. Live usage is shown in Dashboard → Usage. If you hit a limit, Helix will pause and tell you exactly which quota was exceeded.
How Helix handles secrets and env vars
+
Helix can read environment variables that are already configured in your project or sandbox, but it cannot exfiltrate them. It will not write secrets into code unless you explicitly paste them. Treat secrets as normal: keep them in env vars, never in commits.
What Helix can do without asking
+
Read-only operations, safe test runs, and file reads never ask. Write actions, deployments, resource creation, and destructive commands pause for approval unless you set autonomy to level 4 and explicitly allow self-driving mode.
GitHub App permissions
+
The Helix GitHub App needs read access to code and metadata, and write access to pull requests and contents if you want Helix to open PRs and push branches.
Audit log and accountability
+
Every Helix tool call is stored in the thread history. Team owners can review the full transcript, including the prompt, reasoning, commands run, and files changed.
How Helix usage is counted
+
Helix meters three things: queries (user messages and internal turns), sandbox compute seconds, and provider API calls. Each is tracked per day, week, and billing month against your plan limits.
Upgrading or downgrading your plan
+
Visit /pricing or Dashboard → Billing to change your plan. Upgrades apply immediately; downgrades apply at the end of the current billing period.
Usage reset dates
+
Daily counters reset at midnight UTC. Weekly counters reset on Monday UTC. Monthly counters reset on your billing anniversary.
Refunds and billing disputes
+
Cancelling stops auto-renewal at period end. If you were charged after cancelling, email support@launchverse.app with the transaction reference.
Why did Helix stop with a context error?
+
Very long tool output can exceed the model's context window. Helix truncates and summarizes output when possible, but some huge files still need to be offloaded to disk. If you see this error, ask Helix to read only the relevant part of the file next time.
How Helix recovers from failed actions
+
When a tool fails, Helix sees the error as an observation and plans a recovery. This can be retrying with a corrected command, installing a missing dependency, or asking you for clarification. It does not give up on the first failure.
Rate limits and retries
+
Helix backs off automatically when it hits provider rate limits. For Groq, OpenAI, or other model providers, Helix uses exponential backoff. Sandbox commands have a per-call timeout but can be retried if the failure looks transient.
Response time SLA
First-response targets by plan. Times are measured from when the ticket lands in our queue to the first human reply.
| Plan | First response | Channels |
|---|---|---|
| Free | Best effort — community and docs | Docs + email |
| Pro | Within 8 business hours | Email + chat |
| Team | Within 4 business hours | Email + chat |
| Enterprise | Within 1 hour, 24/7 | Email + chat + phone |
Critical issue?
If the Helix brain, dashboard, or deployment flow is fully down, mark your ticket [P1]in the subject line and we'll page on-call. Team and Enterprise customers can also use the in-dashboard Emergency contact button.
Always check status.launchverse.app first — if there's an active platform incident, you'll see it there before you fill out a ticket.