Memories

Memories are a separate archive object from conversation sessions. Hansard keeps the current provider memory content together with full-copy snapshots written when that content changes.

Back up memories

Run a backup pass explicitly:

hansard memory backup

The backup scans supported provider stores, refreshes the memory search index, and records unchanged items without creating duplicate content snapshots.

Supported sources include:

Claude Code project memory and global memory files.
Codex memories, global instructions, and rollout summaries.
Historical Gemini CLI memory instructions.
Antigravity knowledge and implicit stores.
Grok Build global, workspace, and session-log memories.
Prime Agent global continual-harness state and refinement history.
Windsurf global rules and local protobuf memory records.
Historical Cursor pending memories and user rules.
Claude.ai memories included in imported exports.
Devin cloud knowledge when an API key is configured.

Enable read-only agent retrieval

Agent access is deliberately separate from memory backup and ordinary conversation recall. Opt in for a trusted, current integration target:

hansard integrations add-to codex --memories

Replace codex with claude, antigravity, devin, opencode, or grok. The generated integration adds a targeted memory workflow and enables two read-only MCP tools:

search_memories searches memories relevant to a focused query. For project work, it considers both the current project scope and global memory instead of treating provider stores as isolated silos.
get_memory retrieves a bounded content window from a selected record. Agents can page deliberately when more context is needed instead of loading the complete memory lake into every prompt.

Neither tool changes the source provider’s memory. Memory reconciliation and edits remain explicit, user-approved workflows. Installing without --memories leaves the memory tools unavailable, and ordinary recall continues to search conversations only. The installed workflow lets an agent search proactively when a task depends on prior preferences, provider capabilities, durable project conventions, or earlier decisions.

The legacy Cursor CLI and Gemini CLI remain historical import sources, not live integration targets. Grok Build and Antigravity are the current successor targets for those integration families.

Browse in the viewer

Open the Memories view in hansard app or hansard web. The rail groups items by project or global scope, while the detail pane shows current content, metadata, and snapshot history.

Use the rail search control or / to search all current memory content and metadata. Use Cmd/Ctrl+F to find text only inside the currently selected memory. Historical snapshots remain available in the detail view but are not included in archive-wide memory search.

When origin metadata is available, the viewer links a memory to its source conversation and links that conversation back to the memory.

Search from the terminal

hansard memory search "deployment preference"
hansard memory search "project convention" \
  --provider codex \
  --repo github.com/acme/widgets \
  --limit 10

Search can filter by exact provider, repository, or origin:

hansard memory list --origin file --json
hansard memory search "review checklist" --origin provider-cloud

The accepted origin values are file, provider-cloud, and manual-import. JSON search output is a result array with excerpts and scores.

Show or export records

Inspect one record:

hansard memory show <memory-id> --json

Export all memory records with their current content for downstream tooling:

hansard memory export

Import ChatGPT memories

ChatGPT does not include memories in its normal export. Copy the contents from the platform’s memory management page into a local text file, then import it:

hansard memory import chatgpt ./chatgpt-memories.txt --account example-account

File origin
Cloud or manual origin

File-backed items can be editable when Hansard can safely write the live provider file before taking a new snapshot.