openclaw/docs/install/migrating-hermes.md

summary, read_when, title

summary	read_when	title
Move from Hermes to OpenClaw with a previewed, reversible import	You are coming from Hermes and want to keep your model config, prompts, memory, and skills You want to know what OpenClaw imports automatically and what stays archive-only You need a clean, scripted migration path (CI, fresh laptop, automation)	Migrating from Hermes

The bundled Hermes migration provider follows HERMES_HOME and the active Hermes profile, falling back to ~/.hermes on macOS/Linux or %LOCALAPPDATA%\hermes on Windows. It previews every change before applying and redacts secrets in plans and reports. Standalone openclaw migrate writes a verified backup; the fresh onboarding path stages config, credentials, and files and publishes them only after imported inference verifies. An explicit --from path always wins.

Imports require a fresh OpenClaw setup. If you already have local OpenClaw state, reset config, credentials, sessions, and the workspace first, or use `openclaw migrate apply hermes` directly with `--overwrite` after reviewing the plan.

Two ways to import

Detects the active Hermes home/profile and shows a preview before applying.

```bash
openclaw onboard --flow import
```

Or point at a specific source:

```bash
openclaw onboard --import-from hermes --import-source ~/.hermes
```

Use `openclaw migrate` for scripted or repeatable runs. See [`openclaw migrate`](/cli/migrate) for the full reference.

```bash
openclaw migrate hermes --dry-run    # preview only
openclaw migrate apply hermes --yes  # apply with confirmation skipped
```

Add `--from <path>` to override Hermes home/profile discovery.

What gets imported

- Default model selection from Hermes `config.yaml`. With `--agent `, the imported model belongs to that agent; shared defaults and other agents keep their models. - Configured model providers and custom endpoints from `model`, `providers`, and `custom_providers`, including Hermes transport aliases, camelCase provider fields, and model lists with per-model metadata. MCP server definitions from `mcp_servers` or `mcp.servers`, including disabled state, timeouts, parallel-tool support, OAuth scope, compatible TLS fields, and native/resource/prompt tool policy. Literal environment variables and headers require credential-import consent. Hermes-only lifecycle, sampling, elicitation, preflight, keepalive, CA-bundle, password-protected client-key, and pre-registered OAuth-client settings become manual-review items instead of invalid OpenClaw config.

An empty `tools.include` keeps native tools disabled while preserving the resource and prompt utility settings. OpenClaw tool filters support exact names and `*`; Hermes `?` and bracket patterns need manual review. Unsupported include patterns are omitted, and a server with unsupported exclusion patterns is imported disabled until you replace its filter and enable it.

- `SOUL.md` and `AGENTS.md` are copied into the OpenClaw agent workspace. - `memories/MEMORY.md` and `memories/USER.md` are **appended** to the matching OpenClaw memory files instead of overwriting them. - Memory-only surfaces behave differently: the onboarding memory page and the Control UI Memory import page copy these two files under `memory/imports/hermes/` for indexed recall and leave existing workspace memory untouched. Memory config defaults for OpenClaw file memory. External memory providers such as Honcho are recorded as archive or manual-review items so you can move them deliberately. Skills with a `SKILL.md` file under active directories in `skills/` are discovered recursively, flattened into the OpenClaw workspace skill directory, and copied with their support files. Per-skill config values from `skills.config` and global disabled state from `skills.disabled` are preserved. With `--skill`, only the selected skills' config and disabled state are imported. Only the organization mirror selected by `_org/.active_org` is imported. Interactive `openclaw migrate` asks before importing auth credentials, with yes selected by default. Accepted imports include current Hermes OpenAI Codex OAuth entries, OpenCode OpenAI OAuth and GitHub Copilot entries, and the [supported Hermes `.env` keys](/cli/migrate#supported-env-keys). Use `--include-secrets` for non-interactive import, `--no-auth-credentials` to skip credentials, or onboarding's `--import-secrets` flag. After importing Hermes OAuth, do not keep Hermes and OpenClaw using the same refresh grant; reauthenticate one side before running both.

What stays archive-only

The provider copies these into the migration report directory for manual review, but does not load them into live OpenClaw config or credentials:

• plugins/
• sessions/
• logs/
• cron/
• mcp-tokens/
• plans/, workspace/, skins/, and kanban/
• pairing/ and platforms/ stores, plus gateway routing/process state
• state.db, hermes_state.db, projects.db, response_store.db, memory_store.db, verification_evidence.db, kanban.db, and retaindb_queue.db

OpenClaw refuses to execute or trust this state automatically because formats and trust assumptions can drift between systems. Move what you need by hand after reviewing the archive.

Recommended flow

```bash openclaw migrate hermes --dry-run ```

The plan lists everything that will change, including conflicts, skipped items, and sensitive items. Nested secret-looking keys are redacted in the output.

```bash openclaw migrate apply hermes --yes ```

OpenClaw creates and verifies a backup before applying. This non-interactive example imports non-secret state only. Run without `--yes` to answer the credential prompt interactively, or add `--include-secrets` to include supported credentials in an unattended run.

```bash openclaw doctor ```

[Doctor](/gateway/doctor) reapplies any pending config migrations and checks for issues introduced during the import.

```bash openclaw gateway restart openclaw status ```

Confirm the gateway is healthy and your imported model, memory, and skills are loaded.

Conflict handling

Apply refuses to continue when the plan reports conflicts (a file or config value already exists at the target).

Rerun with `--overwrite` only when replacing the existing target is intentional. Providers may still write item-level backups for overwritten files in the migration report directory.

Conflicts are unusual on a fresh install. They typically show up when you re-run the import against a setup that already has user edits.

If a conflict surfaces mid-apply (for example, an unexpected race on a config file), that item is reported as a conflict while independent files, skills, credentials, archives, and config entries continue. Resolve the conflicted item and rerun the import; identical memory imports are idempotent.

Secrets

Interactive openclaw migrate asks whether to import detected auth credentials, with yes selected by default.

• Accepting imports current Hermes OpenAI Codex OAuth entries, OpenCode OpenAI OAuth and GitHub Copilot entries, and the supported .env keys.
• Use --no-auth-credentials, or answer no at the prompt, to import non-secret state only.
• Use --include-secrets to import credentials in an unattended --yes run.
• Use the onboarding wizard's --import-secrets flag to import credentials from the wizard.

JSON output for automation

openclaw migrate hermes --dry-run --json
openclaw migrate apply hermes --json --yes

openclaw migrate hermes --json without --yes prints the plan without applying it. Non-interactive migrate apply requires --yes. A partial apply failure returns the complete JSON report and exits with code 1.

Troubleshooting

Inspect the plan output. Each conflict identifies the source path and the existing target. Decide per item whether to skip, edit the target, or rerun with `--overwrite`. Pass `--from /actual/path` (CLI) or `--import-source /actual/path` (onboarding). Onboarding imports require a fresh setup. Either reset state and re-onboard, or use `openclaw migrate apply hermes` directly, which supports `--overwrite` and explicit backup control. Interactive `openclaw migrate` imports API keys only when you accept the credential prompt. Non-interactive `--yes` runs need `--include-secrets`; onboarding imports need `--import-secrets`. Only the [supported `.env` keys](/cli/migrate#supported-env-keys) are recognized — other `.env` variables are ignored.

Related

• openclaw migrate: full CLI reference, plugin contract, and JSON shapes.
• Onboarding: wizard flow and non-interactive flags.
• Migrating: move an OpenClaw install between machines.
• Doctor: post-migration health check.
• Agent workspace: where SOUL.md, AGENTS.md, and memory files live.