Host Capacity

rensei host capacity controls the execution envelope for this machine: how many agent sessions the daemon runs concurrently, how much disk the workarea pool may consume, and when idle pool members are evicted.

These settings govern this host only. For org-level capacity (execution providers and pools that span all your machines), see rensei capacity.

Capacity Status

Show the current capacity envelope and live session counts.

rensei host capacity status
rensei host capacity status --pool   # include workarea pool breakdown
rensei host capacity status --json

Sample output:

  Active sessions:  2 / 8
  Queue depth:      0
  Max vCPU/session: 4
  Max mem/session:  4096 MB
  Reserved vCPU:    2
  Reserved memory:  1024 MB
  Timestamp:        2026-06-02T14:00:00Z

With --pool:

  Workarea pool:
    Total:      12
    Ready:      8
    Acquired:   2
    Warming:    2
    Invalid:    0
    Disk usage: 18432 MB

    By (repo, toolchain):
    REPO                             TOOLCHAIN  TOTAL  READY  ACQUIRED  WARMING  INVALID  DISK MB
    https://github.com/acme/api.git  node-22    6      4      1         1        0        9216
    https://github.com/acme/web.git  node-22    6      4      1         1        0        9216

Set Capacity Parameters

Update capacity parameters without restarting the daemon. Changes are written atomically to ~/.rensei/daemon.yaml and the daemon reloads the affected subsystem immediately.

rensei host capacity set capacity.maxConcurrentSessions 8
rensei host capacity set capacity.poolMaxDiskGb 50

Supported keys:

Key	Type	Description
`capacity.maxConcurrentSessions`	integer ≥ 0	Maximum concurrently accepted sessions. `0` stops accepting new sessions (pause).
`capacity.poolMaxDiskGb`	integer ≥ 0	Maximum total pool disk in GiB before LRU eviction triggers. `0` means no limit.

Setting maxConcurrentSessions to 0 is equivalent to rensei host pause - new sessions are rejected, in-flight sessions continue.

# Temporarily drop concurrency to 2 during business hours
rensei host capacity set capacity.maxConcurrentSessions 2

# Raise it back to full capacity
rensei host capacity set capacity.maxConcurrentSessions 8

# Cap pool disk at 100 GiB
rensei host capacity set capacity.poolMaxDiskGb 100

If the daemon is unreachable when you run set, the change is written to daemon.yaml on disk and takes effect when the daemon next starts.

Scale Sessions

A shorthand to scale capacity.maxConcurrentSessions to a specific value.

rensei host capacity scale 4   # set maxConcurrentSessions to 4

Evict Pool Members

Manually evict workarea pool members for a repository based on age. Only idle (ready) members are evicted; members that are actively running a session (acquired) are never touched.

rensei host capacity evict \
  --repo https://github.com/acme/api.git \
  --older-than 24h

rensei host capacity evict \
  --repo https://github.com/acme/api.git \
  --older-than 48h \
  --json

Flags

Flag	Required	Description
`--repo`	yes	Repository URL to evict pool members for
`--older-than`	yes	Evict members last acquired (or created if never acquired) more than this duration ago. Go duration string, e.g. `24h`, `168h`.
`--json`	-	Emit raw JSON response

Sample output:

evicted 3 pool member(s): eviction scheduled
correlation-id: evt_01hq...

The correlation ID appears in Layer 6 hook events and can be joined in the observability dashboard (rensei observability events query --filter "correlationId=evt_01hq...").

Workarea Pool States

Each pool member (a pre-cloned + warmed workarea) is in one of four states:

State	Meaning
`ready`	Warm and available for immediate dispatch
`warming`	Being cloned/toolchain-installed; not yet dispatchable
`acquired`	Currently running an agent session
`invalid`	Failed to warm; will be garbage-collected

Configuration Reference

The capacity envelope is stored in ~/.rensei/daemon.yaml under the capacity key:

capacity:
  maxConcurrentSessions: 8
  poolMaxDiskGb: 100
  maxVCpuPerSession: 4
  maxMemoryMbPerSession: 4096
  reservedVCpu: 2
  reservedMemoryMb: 1024

maxVCpuPerSession, maxMemoryMbPerSession, reservedVCpu, and reservedMemoryMb are read-only from the CLI and configured directly in the YAML file.

Daemon - Install, start, stop, and monitor the daemon
Workarea - Inspect and manage individual workarea snapshots
Org-level Capacity - Pools and providers spanning your entire fleet
Fleet - View all machines in the org and configure routing