Tools

Tools

help() returns every tool with a one-line summary; help(command='<name>') returns that tool's signature and its full documentation, read live from the code. "Full" is the operative word: a tool's description is sent with every request for the whole session, so the reasoning and worked detail live behind help() and only what a caller needs in order to choose correctly stays in the schema.

The same arithmetic applies to the tool list itself. All 36 schemas cost about 9.5k tokens of every request whether or not the session ever documents a flow or runs a curation pass, so MEMAI_TOOLS names the groups to publish:

MEMAI_TOOLS	tools	~tokens/request
full (default)	36	9.5k
core,curation	30	7.7k
core,diagrams	28	7.7k
core	22	5.9k

core is the reading, writing, editing and linking surface; diagrams is authoring one (get_diagram stays in core — reading a flow is a read); curation is the optimize/dedup pass plus the store-wide settings and purge_memory. Any group implies core. The default publishes everything, and help() documents every tool, naming the ones this process did not load and the group that turns them on.

Writing
note(content, domain, also, tags, session, review_after, source_ref)	Save a fact/decision/finding (type='note')
checkpoint(intent, established, pursuing, open_questions, session, domain, also)	Save work state; fields are free-length
anti_pattern(pattern, why_wrong, instead, domain, also, session)	Save a pitfall to avoid repeating
reasoning(content, domain, also, session)	Save a reasoning trace
handoff(content, domain, also, session)	Leave a note for another agent/session

Reading
pulse(domain)	Warm-up: latest checkpoint, open handoffs/anti-patterns, recent notes, flow titles, scope census (incl. what is overdue for a recheck)
search(query, domain, type, limit)	Hybrid BM25 + vector search, source-annotated
recall(query, domain, limit)	Relevance-ranked recall of note()'d knowledge
list_by_domain(domain, type, limit, subtree)	Recency-ordered, scoped to a path and its subdomains
list_recent(type, domain, limit, subtree)	Recency-ordered, global
timeline(uid, query, before, after, domain, type)	The records written either side of one memory, in created_at order; anchored by uid, or by a query whose top hit becomes the anchor
list_domains()	The domain tree: own/subtree/cross-listed counts and latest activity
get_memory(uid)	Full record: edit history, relations, referencing diagrams
get_relations(uid)	Relations for a memory

Diagrams
diagram(title, nodes, edges, summary, domain, also, session, tags, kind)	Document a routine as a graph
diagram_node(uid, key, label, shape, note, delete)	Add, patch or remove one step
diagram_edge(uid, from_key, to_key, label, delete)	Wire two steps, relabel or remove the wire
diagram_link(uid, node_key, target_uid, relation_type, delete)	Attach a memory to one step
diagram_jump(uid, node_key, peer_uid, peer_node, label, delete)	Continue a step into another flow
diagram_relayout(uid)	Recompute the stored node positions
get_diagram(uid, format)	Read a diagram back (formats above)

Editing and domains
edit_memory(uid, new_content, note, mode, source_ref)	Correct a memory, or mode='append' add to it; the prior version is kept. source_ref repoints it at its source, on its own or with the edit
link_memories(from_uid, to_uid, relation_type, note)	Create a typed relation
set_confidence(uid, confidence)	unverified | confirmed | contradicted
also_domain(uid, domain)	Cross-list a memory into one more path
unfile_domain(uid, domain)	Drop one cross-listing; where it is filed is untouched
get_domain_case() / set_domain_case(mode)	Read/set the domain-casing policy

Curation and deletion
dedup_scan(domain, type, threshold, limit)	Likely-duplicate pairs, for review
optimize_scan(domain, type, since, include_archived, limit, offset, full)	Dump the corpus compactly to plan a curation pass
optimize_stage(suggestions, note)	Stage a batch of suggestions for human review
optimize_runs() / optimize_status(run_id)	What was staged, and what the human applied or rejected
forget(uid, reason, superseded_by)	Soft delete (archive, reversible)
purge_memory(uid, confirm_phrase)	Hard delete, requires "DELETE <uid>"
help(command)	Tool docs read live from the code's docstrings

Writer tool names match the type they store (note() → type='note', reasoning() → type='reasoning', ...), so the verb an agent calls is exactly the string it later filters on.