How Azhi compares
Feature matrix
Section titled “Feature matrix”| Azhi Flow | OpenCode | VS Code + Copilot agent mode | Claude Code / Agent SDK | n8n-style automation | LangGraph-style frameworks | |
|---|---|---|---|---|---|---|
| Primary shape | Self-hosted workflow runtime | Terminal/coding agent | Agent inside the editor | Coding agent (CLI) and SDK | Visual automation with AI nodes | Code library for agent graphs |
| Unit of work | Workflow (agent is one node) | Agent session | Chat/agent session | Agent session | Workflow | Graph in code |
| Durable runs, retries, schedules | Yes (Temporal) | No (interactive sessions) | No | No built-in scheduler | Yes | Checkpointing; scheduling depends on deployment |
| Many model providers | Anthropic, OpenAI, Copilot today | Yes, many | Models offered by your Copilot plan | Claude models | Many, via nodes | Many |
| MCP | stdio and Streamable HTTP via a governed gateway | Yes | Yes | Yes | Available | Via adapters |
| Pre-run plan of what is enforced vs not | Yes | No | No | No | No | No |
| Exactly-once writes ledger | Yes | No | No | No | No | No |
| Taint check for injected data | Yes, at compile time | No | No | No | No | No |
| Per-turn context manifest | Yes | No | No | No | No | No |
| Cost per step, budgets, spend limits | Yes | Usage per session | Plan-level premium usage | Usage and cost reporting | Varies | Via separate tracing tools |
| Human approval gates | Yes | Per-tool permission prompts | Per-tool confirmation | Per-tool permission prompts | Yes | Yes (interrupts) |
| Best at | Repeatable, governed team workflows | Interactive coding, any model | Coding where you already work | Interactive and headless coding | Integrations glue | Custom agent apps in code |
When to choose which
Section titled “When to choose which”| If you need… | Choose |
|---|---|
| Interactive coding in the editor you already use | VS Code agent mode (or your editor’s assistant) |
| Interactive terminal coding with any model | OpenCode |
| Interactive or headless coding with Claude | Claude Code / Agent SDK |
| Hundreds of SaaS integrations with light AI | n8n-style automation |
| A custom agent application written in code | A LangGraph-style framework |
| The same agent work run repeatedly, unattended, auditable, with limits and sign-offs | Azhi Flow |
How Azhi fits with them
Section titled “How Azhi fits with them”- OpenCode and the Claude Agent SDK are Azhi executors today. Azhi adds what they leave to you: durability, scheduling, approvals, the action ledger, taint checks and cost limits.
- VS Code agent mode is right for a developer working interactively. Azhi is for the same kind of work run repeatedly, unattended and auditable: a weekly report, every PR review, a ticket-to-release flow.
- n8n-style tools are strong on integrations. Azhi’s difference is treating agents as governed, measured nodes with explicit guarantees.
- LangGraph-style frameworks give building blocks in code. Azhi is an opinionated runtime with a declarative definition, compile-time checks and an operations UI.
Honest weak spots
Section titled “Honest weak spots”- Fewer integrations and providers than the broad automation tools and frameworks.
- Linux-only workers and a heavier install than a CLI you
npm i -g. - Capability-based model routing, result caching and savings reports are Vision, not built. See Project status.