diff --git a/DESCRIPTION b/DESCRIPTION
index ee4e91f..ea15be7 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -26,9 +26,8 @@ Imports:
     sf
 Suggests:
     bookdown,
-    furrr,
-    future,
     gq,
+    mirai,
     knitr,
     rmarkdown,
     mockery,
diff --git a/R/frs_db_conn.R b/R/frs_db_conn.R
index 9eccb31..19b0970 100644
--- a/R/frs_db_conn.R
+++ b/R/frs_db_conn.R
@@ -42,3 +42,32 @@ frs_db_conn <- function(
     password = password
   )
 }
+
+
+#' Extract connection parameters from an existing connection
+#'
+#' Reads host, port, dbname, user from a live [RPostgres::Postgres()]
+#' connection. Used by [frs_habitat()] to pass connection params to
+#' parallel workers so they can open their own connections without
+#' depending on `PG_*_SHARE` environment variables.
+#'
+#' Password is not available from `DBI::dbGetInfo()` (security). Must
+#' be provided explicitly via `password` param or the connection will
+#' fail for password-authenticated databases.
+#'
+#' @param conn A [DBI::DBIConnection-class] object.
+#' @param password Character. Password for reconnection. Required for
+#'   password-authenticated databases. Not needed for trust auth or
+#'   `.pgpass` file.
+#' @return Named list with `dbname`, `host`, `port`, `user`, `password`.
+#' @noRd
+.frs_conn_params <- function(conn, password = "") {
+  info <- DBI::dbGetInfo(conn)
+  list(
+    dbname = info$dbname,
+    host = info$host,
+    port = info$port,
+    user = info$username,
+    password = password
+  )
+}
diff --git a/R/frs_habitat.R b/R/frs_habitat.R
index b4d6ff2..75540b1 100644
--- a/R/frs_habitat.R
+++ b/R/frs_habitat.R
@@ -4,7 +4,7 @@
 #' more watershed groups. Calls [frs_habitat_partition()] per WSG to extract
 #' the base network and pre-compute breaks, then flattens all (WSG, species)
 #' pairs and classifies them via [frs_habitat_species()]. Both phases
-#' parallelize with [furrr::future_map()] when `workers > 1`.
+#' parallelize with [mirai::mirai_map()] when `workers > 1`.
 #'
 #' Output tables are WSG-scoped: `working.streams_bulk_co`,
 #' `working.streams_morr_bt`, etc.
@@ -13,13 +13,17 @@
 #' @param wsg Character. One or more watershed group codes
 #'   (e.g. `"BULK"`, `c("BULK", "MORR")`).
 #' @param workers Integer. Number of parallel workers. Default `1`
-#'   (sequential). Values > 1 require the `furrr` package. Each worker
-#'   opens its own database connection. Used for both Phase 1 (partition
-#'   prep across WSGs) and Phase 2 (species classification).
+#'   (sequential). Values > 1 require the `mirai` package. Each worker
+#'   opens its own database connection (params extracted from `conn`).
+#'   Used for both Phase 1 (partition prep across WSGs) and Phase 2
+#'   (species classification).
 #' @param break_sources List of break source specs passed to
 #'   [frs_habitat_access()], or `NULL` for gradient-only. Each spec is a
 #'   list with `table`, and optionally `where`, `label`, `label_col`,
 #'   `label_map`. See [frs_habitat_access()] for details.
+#' @param password Character. Database password for parallel workers.
+#'   Required when `workers > 1` and the database uses password auth.
+#'   Not needed for trust auth or `.pgpass`.
 #' @param cleanup Logical. Drop intermediate tables (base network, break
 #'   tables) when done. Default `TRUE`.
 #' @param verbose Logical. Print progress and timing. Default `TRUE`.
@@ -58,6 +62,7 @@
 #' }
 frs_habitat <- function(conn, wsg, workers = 1L,
                         break_sources = NULL,
+                        password = "",
                         cleanup = TRUE, verbose = TRUE) {
   stopifnot(is.character(wsg), length(wsg) > 0)
 
@@ -106,34 +111,51 @@ frs_habitat <- function(conn, wsg, workers = 1L,
   # Phase 1: Prepare partitions (extract base, pre-compute breaks)
   # ==========================================================================
   workers <- as.integer(workers)
-  use_furrr <- workers > 1L
-  if (use_furrr && !requireNamespace("furrr", quietly = TRUE)) {
-    stop("furrr package required for parallel execution (workers > 1)",
+  use_parallel <- workers > 1L
+  if (use_parallel && !requireNamespace("mirai", quietly = TRUE)) {
+    stop("mirai package required for parallel execution (workers > 1)",
          call. = FALSE)
   }
 
-  .run_partition <- function(spec) {
-    p_conn <- if (use_furrr) frs_db_conn() else conn
-    if (use_furrr) on.exit(DBI::dbDisconnect(p_conn))
-    frs_habitat_partition(p_conn,
-      aoi = spec$aoi,
-      label = spec$label,
-      species = spec$species,
-      params_all = spec$params_all,
-      params_fresh = spec$params_fresh,
-      break_sources = spec$break_sources,
-      verbose = verbose && !use_furrr)
-  }
+  # Extract connection params from user's conn for worker reconnection
+  conn_params <- if (use_parallel) .frs_conn_params(conn, password) else NULL
 
-  if (use_furrr) {
+  if (use_parallel) {
     if (verbose) cat("\nPhase 1: preparing ", length(wsg_specs),
                      " partition(s) (", workers, " workers)...\n", sep = "")
-    old_plan <- future::plan(future::multisession, workers = workers)
-    on.exit(future::plan(old_plan), add = TRUE)
-    partitions <- furrr::future_map(wsg_specs, .run_partition,
-      .options = furrr::furrr_options(seed = TRUE, packages = "fresh"))
+    mirai::daemons(workers)
+    on.exit(mirai::daemons(0), add = TRUE)
+    partitions <- mirai::mirai_map(wsg_specs, function(spec) {
+      library(fresh)
+      p_conn <- do.call(DBI::dbConnect,
+        c(list(drv = RPostgres::Postgres()), conn_params))
+      on.exit(DBI::dbDisconnect(p_conn))
+      frs_habitat_partition(p_conn,
+        aoi = spec$aoi, label = spec$label,
+        species = spec$species, params_all = spec$params_all,
+        params_fresh = spec$params_fresh,
+        break_sources = spec$break_sources, verbose = FALSE)
+    }, conn_params = conn_params)[]
   } else {
-    partitions <- lapply(wsg_specs, .run_partition)
+    partitions <- lapply(wsg_specs, function(spec) {
+      frs_habitat_partition(conn,
+        aoi = spec$aoi, label = spec$label,
+        species = spec$species, params_all = spec$params_all,
+        params_fresh = spec$params_fresh,
+        break_sources = spec$break_sources, verbose = verbose)
+    })
+  }
+
+  # Check for worker errors
+  if (use_parallel) {
+    errs <- vapply(partitions, inherits, logical(1), "miraiError")
+    if (any(errs)) {
+      msgs <- vapply(which(errs), function(i) {
+        paste0(wsg_specs[[i]]$wsg, ": ", conditionMessage(partitions[[i]]))
+      }, character(1))
+      stop("Partition failed:\n  ", paste(msgs, collapse = "\n  "),
+           call. = FALSE)
+    }
   }
 
   # Collect all jobs and cleanup tables
@@ -148,36 +170,56 @@ frs_habitat <- function(conn, wsg, workers = 1L,
   # ==========================================================================
   # Phase 2: Classify all (partition, species) pairs
   # ==========================================================================
-  .run_one <- function(job) {
-    worker_conn <- if (use_furrr) frs_db_conn() else conn
-    if (use_furrr) on.exit(DBI::dbDisconnect(worker_conn))
+  .run_species <- function(job) {
+    library(fresh)
+    worker_conn <- do.call(DBI::dbConnect,
+      c(list(drv = RPostgres::Postgres()), conn_params))
+    on.exit(DBI::dbDisconnect(worker_conn))
     t0 <- proc.time()
     frs_habitat_species(worker_conn, job$species_code, job$base_tbl,
-      breaks = job$acc_tbl,
-      breaks_habitat = job$hab_tbl,
-      params_sp = job$params_sp,
-      fresh_sp = job$fresh_sp,
+      breaks = job$acc_tbl, breaks_habitat = job$hab_tbl,
+      params_sp = job$params_sp, fresh_sp = job$fresh_sp,
       to = job$to)
     elapsed <- (proc.time() - t0)["elapsed"]
-    data.frame(
-      partition = job$partition,
-      species_code = job$species_code,
+    data.frame(partition = job$partition, species_code = job$species_code,
       access_threshold = job$access_threshold,
       habitat_threshold = job$habitat_threshold,
-      elapsed_s = elapsed,
-      table_name = job$to,
-      stringsAsFactors = FALSE
-    )
+      elapsed_s = elapsed, table_name = job$to,
+      stringsAsFactors = FALSE)
+  }
+
+  .run_species_local <- function(job) {
+    t0 <- proc.time()
+    frs_habitat_species(conn, job$species_code, job$base_tbl,
+      breaks = job$acc_tbl, breaks_habitat = job$hab_tbl,
+      params_sp = job$params_sp, fresh_sp = job$fresh_sp,
+      to = job$to)
+    elapsed <- (proc.time() - t0)["elapsed"]
+    data.frame(partition = job$partition, species_code = job$species_code,
+      access_threshold = job$access_threshold,
+      habitat_threshold = job$habitat_threshold,
+      elapsed_s = elapsed, table_name = job$to,
+      stringsAsFactors = FALSE)
   }
 
-  if (use_furrr) {
+  if (use_parallel) {
     if (verbose) cat("Phase 2: classifying (", workers, " workers)...\n",
                      sep = "")
-    result_list <- furrr::future_map(all_jobs, .run_one,
-      .options = furrr::furrr_options(seed = TRUE, packages = "fresh"))
+    result_list <- mirai::mirai_map(all_jobs, .run_species,
+      conn_params = conn_params)[]
+    # Check for worker errors
+    errs <- vapply(result_list, inherits, logical(1), "miraiError")
+    if (any(errs)) {
+      msgs <- vapply(which(errs), function(i) {
+        paste0(all_jobs[[i]]$partition, "/", all_jobs[[i]]$species_code,
+               ": ", conditionMessage(result_list[[i]]))
+      }, character(1))
+      stop("Species classification failed:\n  ",
+           paste(msgs, collapse = "\n  "), call. = FALSE)
+    }
   } else {
     result_list <- lapply(all_jobs, function(job) {
-      res <- .run_one(job)
+      res <- .run_species_local(job)
       if (verbose) {
         cat("  ", res$partition, "/", res$species_code, ": ",
             round(res$elapsed_s, 1), "s -> ", res$table_name, "\n", sep = "")
@@ -189,7 +231,7 @@ frs_habitat <- function(conn, wsg, workers = 1L,
   results <- do.call(rbind, result_list)
   rownames(results) <- NULL
 
-  if (verbose && use_furrr) {
+  if (verbose && use_parallel) {
     for (i in seq_len(nrow(results))) {
       cat("  ", results$partition[i], "/", results$species_code[i], ": ",
           round(results$elapsed_s[i], 1), "s -> ",
diff --git a/man/frs_habitat.Rd b/man/frs_habitat.Rd
index 03916f5..f57ab43 100644
--- a/man/frs_habitat.Rd
+++ b/man/frs_habitat.Rd
@@ -9,6 +9,7 @@ frs_habitat(
   wsg,
   workers = 1L,
   break_sources = NULL,
+  password = "",
   cleanup = TRUE,
   verbose = TRUE
 )
@@ -20,15 +21,20 @@ frs_habitat(
 (e.g. \code{"BULK"}, \code{c("BULK", "MORR")}).}
 
 \item{workers}{Integer. Number of parallel workers. Default \code{1}
-(sequential). Values > 1 require the \code{furrr} package. Each worker
-opens its own database connection. Used for both Phase 1 (partition
-prep across WSGs) and Phase 2 (species classification).}
+(sequential). Values > 1 require the \code{mirai} package. Each worker
+opens its own database connection (params extracted from \code{conn}).
+Used for both Phase 1 (partition prep across WSGs) and Phase 2
+(species classification).}
 
 \item{break_sources}{List of break source specs passed to
 \code{\link[=frs_habitat_access]{frs_habitat_access()}}, or \code{NULL} for gradient-only. Each spec is a
 list with \code{table}, and optionally \code{where}, \code{label}, \code{label_col},
 \code{label_map}. See \code{\link[=frs_habitat_access]{frs_habitat_access()}} for details.}
 
+\item{password}{Character. Database password for parallel workers.
+Required when \code{workers > 1} and the database uses password auth.
+Not needed for trust auth or \code{.pgpass}.}
+
 \item{cleanup}{Logical. Drop intermediate tables (base network, break
 tables) when done. Default \code{TRUE}.}
 
@@ -44,7 +50,7 @@ Orchestrate the full habitat pipeline for all species present in one or
 more watershed groups. Calls \code{\link[=frs_habitat_partition]{frs_habitat_partition()}} per WSG to extract
 the base network and pre-compute breaks, then flattens all (WSG, species)
 pairs and classifies them via \code{\link[=frs_habitat_species]{frs_habitat_species()}}. Both phases
-parallelize with \code{\link[furrr:future_map]{furrr::future_map()}} when \code{workers > 1}.
+parallelize with \code{\link[mirai:mirai_map]{mirai::mirai_map()}} when \code{workers > 1}.
 }
 \details{
 Output tables are WSG-scoped: \code{working.streams_bulk_co},
diff --git a/scripts/habitat/logs/20260404_habitat_benchmark-mirai-parallel.txt b/scripts/habitat/logs/20260404_habitat_benchmark-mirai-parallel.txt
new file mode 100644
index 0000000..66823d1
--- /dev/null
+++ b/scripts/habitat/logs/20260404_habitat_benchmark-mirai-parallel.txt
@@ -0,0 +1,36 @@
+## Habitat pipeline benchmark — mirai parallel
+## Date: 2026-04-04
+## DB: local Docker fwapg (port 5432, 32GB shared_buffers)
+## Branch: mirai-parallel
+## Issue: #77
+
+## Setup
+# WSGs: ADMS (11,520 segments, 5 species) + ATNA (18,684 segments, 7 species)
+# Break sources: falls (barrier_ind = TRUE, label = "blocked")
+# Workers: 4 (mirai daemons)
+# Optimizations: ltree-enriched breaks, split NOT EXISTS, indexes
+
+## Results
+# Phase 1 (partition prep, parallel): ~25s (2 WSGs on 2 workers)
+# Phase 2 (species classification, parallel): ~71s (12 species on 4 workers)
+# Total: 96s
+#
+# Per species:
+#   ADMS: BT 22s, CH 21s, CO 21s, RB 22s, SK 24s
+#   ATNA: CH 15s, CM 14s, CO 15s, PK 17s, RB 17s, SK 17s, ST 13s
+
+## Comparison
+# v0.6.0 sequential (ADMS only, 5 species, remote DB): 3,755s (63 min)
+# v0.7.0 optimized sequential (ADMS only, 5 species): 86s
+# mirai parallel (ADMS + ATNA, 12 species, 4 workers): 96s
+#
+# Effective throughput:
+#   Old: 1 species / 750s = 0.08 species/min
+#   New: 12 species / 96s = 7.5 species/min
+#   Speedup: ~94x throughput improvement
+
+## Notes
+# - mirai replaces furrr/future — lighter daemons, no serialization overhead
+# - Connection params extracted from user's conn, password passed explicitly
+# - Workers load fresh package via library(fresh) — must be installed
+# - Phase 2 is wall-clock ~71s for 12 species on 4 workers (3 rounds of 4)