OpenClaw

OpenClaw is a personal-agent harness maintained by the OpenClaw Foundation. Hansard imports history from the Gateway host. Routed models retain their own model-company identity in Stats.

Source IDs

Source	Use it for
openclaw	Locally available Gateway stores shared by CLI, web, desktop, and messaging clients.

How the import works

OpenClaw saves current conversations in each agent’s agents/<id>/agent/openclaw-agent.sqlite. Hansard reads a coherent snapshot, including committed WAL data, without changing the provider database. It also reads legacy sessions/*.jsonl, retained reset and deleted archives, and verified cold-storage files. When a session has migrated to SQLite, that store owns the conversation and its old JSONL files remain raw preservation sidecars.

The default roots are ~/.openclaw, named ~/.openclaw-<profile> directories, and historical ~/.clawdbot. OPENCLAW_HOME changes the home used for discovery; OPENCLAW_STATE_DIR selects a single custom store. Set HANSARD_OPENCLAW_ROOTS to a path-separated list to import several stores. Each profile and agent keeps its own identity, even if native session IDs match. A resumed conversation updates its existing archive entry without duplicating unchanged sessions.

Clients connected to the same Gateway share its archive source. For a remote or cloud Gateway, make its state directory available locally through a copy or mount. Hansard does not connect to messaging accounts or fetch remote Gateway history. Incognito conversations have no durable store to import.

Import and keep current

Conversation history

hansard import --source openclaw --since all
hansard config sources enable openclaw

The watcher observes agent databases, WAL files, and retained transcript files. Compressed records require Node.js with Zstandard support or zstd on PATH.

Memories and instructions

hansard memory backup

Backups include workspace AGENTS.md, SOUL.md, USER.md, IDENTITY.md, TOOLS.md, HEARTBEAT.md, BOOT.md, BOOTSTRAP.md, MEMORY.md, and daily memory/ text. Skill folders are not memories and are not backed up. Standard workspace directories and workspaces recorded in imported sessions are discovered. OPENCLAW_WORKSPACE_DIR or path-separated HANSARD_OPENCLAW_WORKSPACES adds custom workspaces, including those not yet present in archived conversations.

Review the import in the app

1. When the import finishes, open Conversations and selectFilter & sort.
2. Set Source to OpenClaw and setTime to the window you imported. UseFolder to narrow the list when you use this harness in several projects.
3. Select a conversation to check its messages, tool calls, files, commits, model, and token usage.

To keep archiving new sessions from a local harness, openSettings → Imports. Automaticfollows every supported harness installed on this computer.Custom lets you turn individual harnesses on or off. Cloud sources and account exports never run from here; each needs its own import command. If a conversation you expect is missing after you reset the filters, run the import command on this page again with --explain-skips where the source supports it.

Review tool arguments and numbered edit diffs, then check OpenClaw and its models in Stats. Internal and inter-session prompts do not count as human activity. Scheduled sessions follow the SDK job accounting rules, and delegated children stay outside human conversation-depth samples.

What Hansard preserves

• Prompts, replies, recorded reasoning, inline images, available local media, tools, errors, and edit patches.
• Active conversation branches, compaction and reset markers, titles, timestamps, workspace, and child links.
• Measured input, output, and cache tokens from individual responses, including spent usage on inactive branches.
• Recorded model and thinking changes, catalog-estimated costs, and unknown record counts.
• Coherent raw databases, legacy indexes and transcripts, and referenced cold-storage files.

Copied fork prefixes are excluded when their source ancestor is available locally. Session-level token counters describe current context and are not used as lifetime totals. Missing model or usage data remains missing; selecting a model later does not attribute earlier calls.

Known limitations

• Audited formats are SQLite schema 23 and transcript versions 1–4. Unknown database or transcript versions produce an error rather than a partial replacement.
• Desktop, web, messaging, cloud workers, and live delegation have not received separate native trials. Their shared local storage is covered by the parser; remote stores require local access.
• External-harness mirrors remain OpenClaw conversations. Importing the same work separately from its native harness can duplicate accounting across providers.
• Inline images and recorded media files inside the Gateway media directory or recorded workspace are preserved up to 32 MiB per file and 128 MiB per session. Remote URLs are retained without fetching; unavailable local files remain missing attachments. Unsupported content blocks are counted and retained in raw preservation.
• Custom config includes, plugin-managed memory databases, and unobserved custom workspaces are not independently discovered. Use the workspace override for custom file-backed memories.
• Runtime logs, voice transcription libraries, delivery queues, credentials, and derived memory indexes are not separate conversations or memories. The full raw per-agent SQLite snapshot still preserves its original tables.

Refresh an existing archive

hansard import --source openclaw --since all
hansard memory backup

Original Gateway stores are required for a parser refresh. Import is read-only; Hansard does not perform OpenClaw’s database migration or restore files into it.

Official references: session storage, transcript schema, and workspace memory.