section · guides
What next
From first-success to actually using Construct day-to-day.
You have Construct installed, a project initialized, agents synced, and you've dispatched your first task. Now what.
Things to read once
| If you want to understand... | Read |
|---|---|
| Why Construct works the way it does | Architecture |
| Pick solo / team / enterprise | Deployment model |
| How signals become triaged R&D work | Intake and triage |
| The persona/specialist model | Agents and personas |
| What blocks commits and why | Gates and enforcement |
| The durable-state story | Beads and state |
Things you might want to do
| If you want to... | Go to |
|---|---|
| Add a new specialist or persona to your team | Add a custom agent |
| Inspect what agents are running and why | Inspect running agents |
| Fix a CI failure or a blocked commit | Fix a policy violation |
| Connect to GitHub, Jira, Slack, or Salesforce | Configure providers |
| Swap out the LLM or embedding model | Plug in your own LLM |
| Use Construct conversationally | Connect your editor |
| Check fleet health with Oracle | construct oracle status — see Architecture |
| Run Construct on AWS | Deploy to AWS |
Reference, when you need it
- CLI reference — every command, every flag.
- Configuration — env vars, config files, what each does.
- Hooks — the hooks that fire on file edits, commits, pushes, prompts.
- MCP tools — tools exposed to Claude Code / OpenCode via MCP.
When something breaks
- Run
construct doctorfirst. Almost everything points at a specific fix. - Check the troubleshooting guide for known patterns.
- In
solomode (the default), Construct runs entirely locally. If a cloud API is down, you can still work fromplan.md,.construct/context.md, the latest handoff, beads, git, and the local vector index. See deployment model for team/enterprise options.