Recall

Recall

Find something in the user's archived conversations. The archive spans every agent that has been imported — Codex, Claude Code and pi — so a conversation is findable from whichever one the user is in now.

Everything below is remaimber, a CLI on PATH. If this agent also exposes the remaimber MCP tools (find_context, get_segments, get_session), prefer them for the first two lookups: same data, fewer round trips, and they return structured results.

Hosts namespace MCP tools by server, so they appear as mcp__remaimber__find_context, mcp__remaimber__get_segments and so on. The bare names used below are the short form; if a lookup for the bare name finds nothing, the tool is probably there under the namespaced one.

Pick the right lookup for what they remember

Three lookups exist, and they fail in different ways. Choose by what the user actually said.

They remember roughly what happened, not the wording — "the part where we set up a mail relay on the nas", "when we fixed the flaky tests":

remaimber resume --match 'the part where we set up a mail relay on the nas'

This is the default for a described memory. It searches every conversation and ranks passages — contiguous stretches actually about the topic — returning each with its session, time span and segment summaries. It handles prose, filler words and hyphenation, and it excludes the live session, so the current conversation cannot rank its own discussion of the topic above the one being looked for. The MCP equivalent is find_context with their phrasing as topic.

They remember an exact string — an error message, a flag, a function name:

remaimber search 'exit status 1' --limit 10

Matches literal message text.

They remember an outcome but no phrasing at all — "when did we sort out the DHCPv6 stalls?":

remaimber recall 'dhcpv6 stalls'

Searches segment summaries rather than raw text, so it matches what the work turned out to be even when nobody typed those words together.

Scoping to one agent

The archive spans agents, and the two entry points default differently. The MCP tools (mcp__remaimber__find_context and friends) search this agent's own conversations unless told otherwise — pass agent: "all" to search every agent, or name one (claude, codex, pi). The CLI is the other way round: it searches everything, and --agent <name> narrows it.

So when a lookup comes back empty, widening the agent scope is the first thing to try, not the last: the work may well have happened in another agent.

Filters

All three accept scoping. Add them when the user implies a scope:

• --repo . restricts to the current repo across every worktree; --subpath . narrows to the current sub-project
• --agent claude|codex|pi restricts to one agent (the CLI searches all of them by default)
• --project <name> filters by project key
• --since <date> / --until <date> for date ranges (ISO 8601, e.g. 2026-08-06T11:18)
• --role user / --role assistant (search only)

search skips tool output by default — command and file output is machine noise, and it includes remaimber's own archived results, so leaving it in makes a search match earlier searches for the same term. Pass --include-tool-output only when the user is specifically looking for something a command printed.

Presenting results

Give session ID (first 8 chars), timestamp, project, and the matching snippet. A leading * means resumable. Sessions from another agent are tagged with its name next to the project; mention that when it matters, since resuming one means opening that agent.

Search results carry a segment_seq — mention it when offering to pull that part back, since it feeds straight into remaimber resume <id> --segments <seq>.

If nothing comes back, say so plainly and check remaimber stats for summary coverage before concluding the conversation isn't archived — an unsummarized session is invisible to recall and to segment lookups, though search still finds it.