babel control

pulse

overview of the live proxy

usage · last 24h: effective tokens per bucket

—

Claude Token Pool & quota

—

subscription OAuth — codex · xai

—

model	provider	reqs	out	effective	cache hit%	cold%	cost
—

recent requests

—

green 2xx · amber 4xx · red 5xx/err · hover a bar for detail

status

live service · Claude Token Pool · database · routes · layers · throughput

service

—

Claude Token Pool

—

database

—

routes

—

disguise layers

—

usage · last 24h: effective tokens per bucket

—

recent requests

—

green 2xx · amber 4xx · red 5xx/err · hover a bar for detail

model	provider	reqs	out	effective	cache hit%	cold%	cost
—

ops log: server-side events (failover · degrade · probe · prune · config · maintenance · worker), newest first

—

requests

every proxied call, newest first · click a row for full detail

id	time	route	model	token	status	dur	agent	channel
—

id range	span	route	model	token	#err	#eff tok	agent	channel
—

sessions

first-class client sessions · drill in for the timeline · —

session	reqs	fails	span	models	tokens	out	last activity
—

session

—

timeline

—

tokens

the Claude Token Pool · sticky selection · ● active = the token serving now · activate switches it, pause forces failover

label · token	pri	status	resets / countdown	reqs · last used	actions
—

add a token

token usage

consumption + cost over time · from the request log

throughput · last 24h: effective tokens per bucket

—

by provider

—

by channel: OpenClaw channels; null grouped as (none)

channel	reqs	prompt	out	effective	cost
—

by agent: agent_id role tag; null grouped as (none)

agent	reqs	prompt	out	cached	effective	cache hit%	cost
—

by model

model	provider	reqs	prompt	out	cached	effective	cache hit%	cold%	cost
—

model registry

one item per full provider-prefixed model id · pricing via own fields or pricing_ref sharing · listed advertises the item in GET /v1/models · cost in stats uses the same resolved pricing

id	owned_by	aliases	ctx backend	price in/out $/M or ←ref	listed /v1/models
—

seen in traffic · no registry entry: models used recently with no pricing/entry; click register to prefill

model	requests	last seen	route
nothing unregistered in recent traffic

add / edit an item: any custom metadata lives in extra JSON and folds into /v1/models output untouched

settings

live config · live keys apply instantly · restart-required keys save to config.json (reload to take effect)

config file

—

live-tunable: applies now, no restart

loading…

restart-required: saved to config.json, applies on next boot

loading…

agent labels: display-only (agent_id → alias + icon); null drops the built-in

loading…

prefix routes: auth kind + tokenFile status; literal token paste only for token routes (never displayed)

loading…

read-only keys: env-only / not forwarded

loading…

maintenance

prune session bodies · WAL checkpoint · integrity check

prune now: clears response bodies + non-frontier request bodies for sessions older than the age floor (min 7d)

age days (≥7)

max sessions (0=∞)

auto-prune schedule: in-process timer; persisted to config.json

enabled

interval

age days (≥7)

max sessions (0=∞)

apply (else dry-run only) pre-backup offline-only (checked = REFUSES apply until you take your own offline backup)

auto body wipe: default OFF; when ON, wipes ALL request + response bodies of every finalized request older than the age floor (min 1d). Unlike auto-prune above, NOTHING is kept (no frontier peaks, no session schema needed). Only the six body columns are touched — rows, stats, usage and sessions survive. IRREVERSIBLE without a backup; babel takes no backup (take your own OFFLINE with babel stopped).

enabled

interval

age days (≥1)

WAL checkpoint

mode

pulse

usage · last 24h: effective tokens per bucket

Claude Token Pool & quota

subscription OAuth — codex · xai

recent requests

status

service

Claude Token Pool

database

routes

disguise layers

usage · last 24h: effective tokens per bucket

recent requests

ops log: server-side events (failover · degrade · probe · prune · config · maintenance · worker), newest first

requests

sessions

session

timeline

transcript: reconstructed conversation events

tokens

add a token

token usage

throughput · last 24h: effective tokens per bucket

by provider

by channel: OpenClaw channels; null grouped as (none)

by agent: agent_id role tag; null grouped as (none)

by model

model registry

seen in traffic · no registry entry: models used recently with no pricing/entry; click register to prefill

add / edit an item: any custom metadata lives in extra JSON and folds into /v1/models output untouched

settings

config file

live-tunable: applies now, no restart

restart-required: saved to config.json, applies on next boot

agent labels: display-only (agent_id → alias + icon); null drops the built-in

prefix routes: auth kind + tokenFile status; literal token paste only for token routes (never displayed)

read-only keys: env-only / not forwarded

maintenance

prune now: clears response bodies + non-frontier request bodies for sessions older than the age floor (min 7d)

auto-prune schedule: in-process timer; persisted to config.json

WAL checkpoint

integrity check: read-only; can take ~seconds on a large DB