Archestra Logo
Archestra.AI
2026-07-16

New: Migrating from Claude, OpenClaw, Hermes to centralized platform

The migration kit inventories an existing agentic setup of personal agents and migrates it to a centralized agentic runtime in Archestra for further collaboration in your org.

New: Migrating from Claude, OpenClaw, Hermes to centralized platform

Written by

Joey Orlando

Why to migrate?

Agents running in OpenClaw, Claude Code, Hermes are great unless you want to share them with your colleagues, bring RBAC, manage third-party system access, log everything and restrict those agents with guardrails to avoid data exfiltration.
Before version 1.3 Archestra simply was not able to execute arbitrary code, had no filesystem, and there was no way to port agents from tools that have access to the host machine. Now, once "Lyra" is out with code execution and persistent files, it's capable of doing so — meaning one could directly migrate personal agents, PoCs, and prototypes into the highly-controllable corporate environment of Archestra.

Migration with Skill

The kit ships as a Skill (migrate-to-archestra) for your favorite coding agent, so the migration runs as a guided, agentic flow rather than a one-shot script. The deterministic work — discovering source artifacts, redacting secrets, building and validating API payloads — lives in zero-dependency Python helpers that run on a stock interpreter with no network, so it works the same on a locked-down enterprise host as on a laptop. The model owns the judgment calls: what maps to what, what to skip, what needs review.
Install it next to the source project:
curl -fsSL https://raw.githubusercontent.com/archestra-ai/archestra/main/migration-kit/install.py | python3
Then ask your coding agent to use the migrate-to-archestra skill against your Archestra instance. The flow:
  1. Connect to an Archestra instance — your own, or a local Docker one the skill can start for you.
  2. Discover the source setup into a secret-redacted inventory.json: agent instructions, skills, subagents, slash commands, local tools, MCP servers, hooks, LLM keys. Credentials are never written to the file.
  3. Review the preview plan and confirm the few decisions that matter: personal, team, or org scope; which MCP servers to install; which LLM keys to bring.
  4. Apply — dry-run first, then the approved plan. Re-running is idempotent, so a second pass updates rather than duplicates.
  5. Report — what moved, what was skipped, what failed, and what needs hands-on review.
Each source primitive has a home in Archestra: the root CLAUDE.md becomes the primary agent's system prompt, skills migrate verbatim with their bundled files, subagents and slash commands become skills, MCP servers become private catalog items you can optionally install, and Claude-compatible lifecycle hooks become native Archestra hooks. Guard hooks become tool policies when the target tool exists; anything that can't map cleanly is flagged for manual follow-up rather than silently dropped.

Getting started

The migration kit works best with Archestra 1.3.6 "Lyra" (yes, that's six follow-up releases, closing the small quirks you've been reporting 😅). Install or update to 1.3.6, point the kit at your existing setup, and see your pilot running in Archestra before the coffee gets cold ☕️
The migration docs cover the details, and the full source lives in migration-kit/ on GitHub.