ingest

Import what your coding agents already know. ingest memories brings in the memory files Claude Code and Codex keep on this machine as typed memories. ingest claude-code and ingest codex read exported transcripts as JSONL and stream each turn into Sibyl's raw memory store, where sibyl reflect can later promote it into durable graph memory.

For files, directories, and URLs of arbitrary documents, use sibyl docs; for crawling websites, use sibyl crawl.

Commands

Command	Description
sibyl ingest memories	Import the memory files Claude Code and Codex keep
sibyl ingest claude-code	Import Claude Code transcript JSONL
sibyl ingest codex	Import Codex transcript JSONL

The two transcript subcommands share the same options and the same import pipeline; only the source adapter differs.

---

ingest memories

Import the memory files your coding agents keep on this machine. They are already distilled by the agent, so each file lands whole as one raw memory that carries its kind (rule, note, artifact, session), rather than as transcript turns. Recall finds them right away, and sibyl reflect can promote them into graph memory. The files are read here and sent through the API, so this works against any server, including a team server.

sibyl ingest memories [OPTIONS]

What it reads

Agent	Files	Kind
Claude Code	~/.claude/projects/*/memory/*.md	feedback → rule, reference → artifact, project → note, user → private note
Codex	~/.codex/memories/rollout_summaries/*.md	One session per summary, attached to the repository it ran in
Codex	~/.codex/memories/memory_summary.md	Your Codex user profile, as a private note

Index files (Claude Code's MEMORY.md, Codex's MEMORY.md and raw_memories.md) only point at these and are skipped, as are symlinks, empty files, and files over 256 KiB.

Where memories go

Memories import into the server this CLI talks to (pass -C <context> to pick another), private to you by default. Each one routes the way commands run in its repository would:

• A memory from a repository linked to a project on that server attaches to the project.
• A memory from a repository pinned to another context is left for that server; the summary prints the sibyl -C <context> ingest memories command that imports it.
• Personal memories (Claude Code user memories and the Codex profile) never attach to a project and always stay private.

Re-running

Each file keeps a stable source ID, so running the import again is safe:

• An unchanged file is skipped.
• A changed file revises its memory in place; Sibyl keeps the earlier content in its revision history.
• After you link a repository, its memories import again attached to the project, and the unattached copies are marked superseded.
• A memory you retired in Sibyl (deleted, superseded, archived) is left alone.

Options

Option	Short	Default	Description
--from		both	Agent to import from: claude-code or codex (repeatable)
--share		false	Make memories from linked repositories visible to their project
--dry-run		false	Show what would change without writing
--yes	-y	false	Don't ask before writing
--claude-dir		~/.claude	Claude Code home
--codex-dir		~/.codex	Codex home
--json	-j	false	JSON summary for scripting

--share applies when a memory is first imported; share or promote an existing memory with the memory sharing commands.

Examples

# See what would import, without writing
sibyl ingest memories --dry-run

# Import from Claude Code only
sibyl ingest memories --from claude-code

# Import into the team server, sharing project memories with each project
sibyl -C team ingest memories --share

sibyl setup offers this import once a machine is connected.

---

ingest claude-code

Import a Claude Code transcript JSONL file or a directory of them.

sibyl ingest claude-code <source> [options]

Argument	Required	Description
source	Yes	Claude Code JSONL file or directory

Option	Default	Description
--scope	private	Target memory scope
--scope-key	(none)	Target scope key for project/team/shared scopes
--source-identity	(none)	Stable identity for moved transcript exports
--batch-size	100	Import batch size
--drain	false	Wait for the background drain to finish
--poll-interval	1.0	Seconds between drain status checks
--timeout	(none)	Maximum seconds to wait when draining
--json / -j	false	JSON output for scripting

Examples

# Import a single session export
sibyl ingest claude-code ~/.claude/projects/sibyl/session.jsonl

# Import a directory and wait for it to finish
sibyl ingest claude-code ~/.claude/projects/sibyl/ --drain

# Import into a shared project scope as JSON
sibyl ingest claude-code session.jsonl --scope project --scope-key sibyl --json

---

ingest codex

Import a Codex transcript JSONL file or a directory of them. Identical to claude-code apart from the source format.

sibyl ingest codex <source> [options]

Argument	Required	Description
source	Yes	Codex JSONL file or directory

Option	Default	Description
--scope	private	Target memory scope
--scope-key	(none)	Target scope key for project/team/shared scopes
--source-identity	(none)	Stable identity for moved transcript exports
--batch-size	100	Import batch size
--drain	false	Wait for the background drain to finish
--poll-interval	1.0	Seconds between drain status checks
--timeout	(none)	Maximum seconds to wait when draining
--json / -j	false	JSON output for scripting

Examples

sibyl ingest codex ~/.codex/sessions/today.jsonl
sibyl ingest codex ~/.codex/sessions/ --drain --timeout 120

Notes

• The source must be a real file or directory, not a symlink, and the path is resolved before upload.
• Imports queue by default and drain in the background. Pass --drain to block until the import reaches a terminal state (completed, failed, or canceled).
• Each imported turn lands as a raw memory. Use --source-identity to keep dedupe stable when you move or rename the same export between runs.

Related Commands

• sibyl reflect - Promote raw memories into graph memory
• sibyl context - Recall raw memories with --raw
• sibyl docs - Import files, directories, or URLs as documents
• sibyl crawl - Crawl and ingest documentation sites