===== somm =====
usage: somm [-h] [--version] [--project PROJECT]
            {chatgpt,status,cache-advice,generate,bench,tail,compare,frontier,workload,boundary,prompt,eval,optimize,campaign,inbox,plugin,mcp,skill,doctor,loop,prune,ledgers,fleet,mirror,alias,serve,spend,climb,gpu,backfill-costs,plans,drain-spool} ...

somm — self-hosted LLM telemetry

positional arguments:
  {chatgpt,status,cache-advice,generate,bench,tail,compare,frontier,workload,boundary,prompt,eval,optimize,campaign,inbox,plugin,mcp,skill,doctor,loop,prune,ledgers,fleet,mirror,alias,serve,spend,climb,gpu,backfill-costs,plans,drain-spool}
    chatgpt             Somm-owned ChatGPT subscription connection
    status              show call roll-up for the current project
    cache-advice        find low prefix-cache reuse opportunities
    generate            one-shot LLM call through somm
    bench               run latency and throughput probes
    tail                stream new calls as they land
    compare             run a prompt through N models side-by-side
    frontier            adequacy frontier per (provider, model) for one workload
    workload            register and inspect project workloads
    boundary            data boundary: route tiers, ledger audit, per-workload boundary
    prompt              manage prompt versions, labels, and A/B variants
    eval                promote datasets and run eval gates
    optimize            propose a prompt fork from failing graded calls
    campaign            run durable experiment campaigns
    inbox               list, show, apply, and dismiss recommendations
    plugin              list and inspect plugins, hooks, and providers
    mcp                 register the somm MCP server with coding agents
    skill               link the somm skill into coding agents' skill directories
    doctor              check config + ollama + db + intel + workers + cooldowns
    loop                run or schedule the intelligence loop (intel refresh, shadow grading,
                        recommendations)
    prune               delete calls (and their samples/updates/evals) older than a retention
                        window
    ledgers             every ledger on this machine: who writes it, its size, and the mirror gap
    fleet               per-project fleet health across registered ledgers
    mirror              the cross-project mirror (~/.somm/global.sqlite)
    alias               canonical model identities (list, seed, set)
    serve               run the web admin + HTTP API (localhost:7878)
    spend               today's LLM spend vs daily budget cap per workload
    climb               what a workload answers before any model (reads chipsocket receipts)
    gpu                 rented-GPU spend read from gipper receipts (read-only)
    backfill-costs      recompute cost_usd for $0 calls that now have pricing intel
    plans               metered-plan quota usage + pacing (PAYG vs metered, fleet-wide)
    drain-spool         replay spooled JSONL telemetry (written during DB outages) into the DB

options:
  -h, --help            show this help message and exit
  --version             show program's version number and exit
  --project PROJECT     project name; also accepted after any command (default: $SOMM_PROJECT,
                        else this directory's project)

exit codes:
  0   success
  1   the command ran and the outcome was a failure: a failed call, a bench or
      compare with no successful call, a failing eval or campaign gate, a data-boundary
      violation, an unhealthy `doctor`, a loop worker error, or a missing prerequisite
      (ledger, receipts directory, harness)
  2   bad usage, or something you named does not exist (workload, prompt version,
      recommendation, project)
  With --json, an unhandled error is printed to stderr as {"ok": false, "error": {...}}
  and exits 2 (bad value), 66 (file not found), 77 (permission denied) or 1.
  Details: docs/cli.md

===== somm chatgpt =====
usage: somm chatgpt [-h] {login,status,models,quota} ...

positional arguments:
  {login,status,models,quota}

options:
  -h, --help            show this help message and exit

===== somm chatgpt login =====
usage: somm chatgpt login [-h] [--port PORT] [--no-browser]

options:
  -h, --help    show this help message and exit
  --port PORT
  --no-browser

===== somm chatgpt status =====
usage: somm chatgpt status [-h]

options:
  -h, --help  show this help message and exit

===== somm chatgpt models =====
usage: somm chatgpt models [-h]

options:
  -h, --help  show this help message and exit

===== somm chatgpt quota =====
usage: somm chatgpt quota [-h]

options:
  -h, --help  show this help message and exit

===== somm status =====
usage: somm status [-h] [--project PROJECT] [--since SINCE] [--json] [--global]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --since SINCE      window in days (default 7)
  --json             emit stable JSON
  --global           read from ~/.somm/global.sqlite (requires SOMM_CROSS_PROJECT)

===== somm cache-advice =====
usage: somm cache-advice [-h] [--project PROJECT] [--since SINCE] [--min-tokens-in MIN_TOKENS_IN]
                         [--max-cache-read-ratio MAX_CACHE_READ_RATIO] [--json]

options:
  -h, --help            show this help message and exit
  --project PROJECT
  --since SINCE         window in days (default 7)
  --min-tokens-in MIN_TOKENS_IN
  --max-cache-read-ratio MAX_CACHE_READ_RATIO
  --json                emit stable JSON

===== somm generate =====
usage: somm generate [-h] [--prompt-file PROMPT_FILE] [--image PATH_OR_URL] [--project PROJECT]
                     [--workload WORKLOAD] [--system SYSTEM] [--provider PROVIDER] [--model MODEL]
                     [--max-tokens MAX_TOKENS] [--temperature TEMPERATURE] [--json]
                     [--claude-cli-text-only]
                     [prompt]

positional arguments:
  prompt                prompt text, or '-' to read stdin

options:
  -h, --help            show this help message and exit
  --prompt-file PROMPT_FILE
                        read prompt text from a file
  --image PATH_OR_URL   attach an image (file path or http(s) URL); repeatable. Routed only to
                        providers that accept images.
  --project PROJECT
  --workload WORKLOAD
  --system SYSTEM
  --provider PROVIDER
  --model MODEL
  --max-tokens MAX_TOKENS
  --temperature TEMPERATURE
  --json                emit a machine-readable result
  --claude-cli-text-only
                        disable Claude CLI tools, MCP config, and settings; requires --provider
                        claude-cli

===== somm bench =====
usage: somm bench [-h] [--project PROJECT] {latency,throughput} ...

positional arguments:
  {latency,throughput}
    latency             run a latency benchmark
    throughput          run a throughput benchmark

options:
  -h, --help            show this help message and exit
  --project PROJECT

===== somm bench latency =====
usage: somm bench latency [-h] [--prompt-file PROMPT_FILE] [--project PROJECT]
                          [--workload WORKLOAD] [--provider PROVIDER] [--model MODEL]
                          [--iterations ITERATIONS] [--warmup WARMUP] [--max-tokens MAX_TOKENS]
                          [--temperature TEMPERATURE] [--allow-fallback] [--json]
                          [prompt]

positional arguments:
  prompt                prompt text, or '-' to read stdin

options:
  -h, --help            show this help message and exit
  --prompt-file PROMPT_FILE
                        read prompt text from a file
  --project PROJECT
  --workload WORKLOAD
  --provider PROVIDER
  --model MODEL
  --iterations ITERATIONS
  --warmup WARMUP
  --max-tokens MAX_TOKENS
  --temperature TEMPERATURE
  --allow-fallback      let a failed --provider/--model pin be rescued by the router chain (off by
                        default: a benchmark that silently changes model measures nothing)
  --json                emit a machine-readable result

===== somm bench throughput =====
usage: somm bench throughput [-h] [--prompt-file PROMPT_FILE] [--project PROJECT]
                             [--workload WORKLOAD] [--provider PROVIDER] [--model MODEL]
                             [--iterations ITERATIONS] [--warmup WARMUP] [--max-tokens MAX_TOKENS]
                             [--temperature TEMPERATURE] [--allow-fallback] [--json]
                             [prompt]

positional arguments:
  prompt                prompt text, or '-' to read stdin

options:
  -h, --help            show this help message and exit
  --prompt-file PROMPT_FILE
                        read prompt text from a file
  --project PROJECT
  --workload WORKLOAD
  --provider PROVIDER
  --model MODEL
  --iterations ITERATIONS
  --warmup WARMUP
  --max-tokens MAX_TOKENS
  --temperature TEMPERATURE
  --allow-fallback      let a failed --provider/--model pin be rescued by the router chain (off by
                        default: a benchmark that silently changes model measures nothing)
  --json                emit a machine-readable result

===== somm tail =====
usage: somm tail [-h] [--project PROJECT] [--workload WORKLOAD] [--since-minutes SINCE_MINUTES]
                 [--poll-interval POLL_INTERVAL]

options:
  -h, --help            show this help message and exit
  --project PROJECT
  --workload WORKLOAD   filter to a single workload
  --since-minutes SINCE_MINUTES
                        start from N minutes ago (default: now)
  --poll-interval POLL_INTERVAL

===== somm compare =====
usage: somm compare [-h] --models MODELS [--project PROJECT] [--workload WORKLOAD]
                    [--max-tokens MAX_TOKENS] [--temperature TEMPERATURE] [--truncate TRUNCATE]
                    [--json]
                    prompt

positional arguments:
  prompt                the prompt text

options:
  -h, --help            show this help message and exit
  --models MODELS       provider/model (repeatable or comma-separated)
  --project PROJECT
  --workload WORKLOAD
  --max-tokens MAX_TOKENS
  --temperature TEMPERATURE
  --truncate TRUNCATE   truncate response text at N chars (0 = no truncate)
  --json                emit JSON instead of the table

===== somm frontier =====
usage: somm frontier [-h] --workload WORKLOAD [--project PROJECT] [--since SINCE] [--json]

options:
  -h, --help           show this help message and exit
  --workload WORKLOAD  workload name to inspect
  --project PROJECT
  --since SINCE        window in days (default 30)
  --json               emit JSON instead of the table

===== somm workload =====
usage: somm workload [-h] [--project PROJECT] {add,create,list,show,set-constraints} ...

positional arguments:
  {add,create,list,show,set-constraints}
    add (create)        register a workload in the project DB
    list                list registered workloads
    show                show one registered workload
    set-constraints     update workload adequacy constraints

options:
  -h, --help            show this help message and exit
  --project PROJECT

===== somm workload add =====
usage: somm workload add [-h] [--project PROJECT] [--description DESCRIPTION]
                         [--privacy-class {public,internal,private}]
                         [--from-example {freeform,structured-extraction}]
                         [--max-p95-latency-ms MAX_P95_LATENCY_MS]
                         [--max-p95-ttft-ms MAX_P95_TTFT_MS] [--max-tpot-ms MAX_TPOT_MS]
                         [--max-capability-failure-rate MAX_CAPABILITY_FAILURE_RATE]
                         [--max-cost-per-call-usd MAX_COST_PER_CALL_USD]
                         [--data-boundary {local,zdr,subscribed,unvetted}]
                         name

positional arguments:
  name                  workload name

options:
  -h, --help            show this help message and exit
  --project PROJECT
  --description DESCRIPTION
  --privacy-class {public,internal,private}
  --from-example {freeform,structured-extraction}
                        built-in workload template (default: freeform)
  --max-p95-latency-ms MAX_P95_LATENCY_MS
  --max-p95-ttft-ms MAX_P95_TTFT_MS
  --max-tpot-ms MAX_TPOT_MS
  --max-capability-failure-rate MAX_CAPABILITY_FAILURE_RATE
  --max-cost-per-call-usd MAX_COST_PER_CALL_USD
  --data-boundary {local,zdr,subscribed,unvetted}
                        loosest route tier this workload's data may reach (default: derive)

===== somm workload list =====
usage: somm workload list [-h] [--project PROJECT] [--json]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --json             emit JSON instead of the table

===== somm workload show =====
usage: somm workload show [-h] [--project PROJECT] name

positional arguments:
  name               workload name

options:
  -h, --help         show this help message and exit
  --project PROJECT

===== somm workload set-constraints =====
usage: somm workload set-constraints [-h] [--project PROJECT]
                                     [--max-p95-latency-ms MAX_P95_LATENCY_MS]
                                     [--max-p95-ttft-ms MAX_P95_TTFT_MS]
                                     [--max-tpot-ms MAX_TPOT_MS]
                                     [--max-capability-failure-rate MAX_CAPABILITY_FAILURE_RATE]
                                     [--max-cost-per-call-usd MAX_COST_PER_CALL_USD] [--clear]
                                     name

positional arguments:
  name                  workload name

options:
  -h, --help            show this help message and exit
  --project PROJECT
  --max-p95-latency-ms MAX_P95_LATENCY_MS
  --max-p95-ttft-ms MAX_P95_TTFT_MS
  --max-tpot-ms MAX_TPOT_MS
  --max-capability-failure-rate MAX_CAPABILITY_FAILURE_RATE
  --max-cost-per-call-usd MAX_COST_PER_CALL_USD
  --clear               clear all workload constraints

===== somm boundary =====
usage: somm boundary [-h] [--project PROJECT] {routes,audit,set} ...

positional arguments:
  {routes,audit,set}
    routes            every configured route with its tier and basis
    audit             per workload: declared boundary vs the worst tier that served it
    set               set a workload's data boundary

options:
  -h, --help          show this help message and exit
  --project PROJECT

===== somm boundary routes =====
usage: somm boundary routes [-h] [--project PROJECT] [--json] [route]

positional arguments:
  route              openrouter/<model-id>: each upstream endpoint, judged from live ZDR evidence

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --json

===== somm boundary audit =====
usage: somm boundary audit [-h] [--project PROJECT] [--since SINCE] [--json] [-v]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --since SINCE      window, e.g. 30d, 12h (default 30d)
  --json
  -v, --verbose      list routes per workload

===== somm boundary set =====
usage: somm boundary set [-h] [--project PROJECT] workload {local,zdr,subscribed,unvetted,none}

positional arguments:
  workload
  {local,zdr,subscribed,unvetted,none}
                        tier, or none to derive

options:
  -h, --help            show this help message and exit
  --project PROJECT

===== somm prompt =====
usage: somm prompt [-h] [--project PROJECT] {list,show,register,fork,diff,label,promote,score} ...

positional arguments:
  {list,show,register,fork,diff,label,promote,score}
    list                list prompt versions for a workload
    show                show a prompt version or label
    register            register a new prompt version
    fork                fork a prompt from a version or label
    diff                diff two prompt versions or labels
    label               move a label to a version or weights
    promote             promote a version to a label
    score               show per-version eval rollups

options:
  -h, --help            show this help message and exit
  --project PROJECT

===== somm prompt list =====
usage: somm prompt list [-h] --workload WORKLOAD [--json] [--project PROJECT]

options:
  -h, --help           show this help message and exit
  --workload WORKLOAD
  --json               emit JSON instead of the table
  --project PROJECT

===== somm prompt show =====
usage: somm prompt show [-h] --workload WORKLOAD [--project PROJECT] [--version VERSION |
                        --label LABEL] [--full]

options:
  -h, --help           show this help message and exit
  --workload WORKLOAD
  --project PROJECT
  --version VERSION
  --label LABEL
  --full

===== somm prompt register =====
usage: somm prompt register [-h] --workload WORKLOAD [--project PROJECT] (--body-file BODY_FILE |
                            --body BODY) [--bump {minor,major}]

options:
  -h, --help            show this help message and exit
  --workload WORKLOAD
  --project PROJECT
  --body-file BODY_FILE
  --body BODY
  --bump {minor,major}

===== somm prompt fork =====
usage: somm prompt fork [-h] --workload WORKLOAD [--project PROJECT] --from FROM_REF
                        --body-file BODY_FILE

options:
  -h, --help            show this help message and exit
  --workload WORKLOAD
  --project PROJECT
  --from FROM_REF
  --body-file BODY_FILE

===== somm prompt diff =====
usage: somm prompt diff [-h] --workload WORKLOAD [--project PROJECT] a b

positional arguments:
  a
  b

options:
  -h, --help           show this help message and exit
  --workload WORKLOAD
  --project PROJECT

===== somm prompt label =====
usage: somm prompt label [-h] --workload WORKLOAD [--project PROJECT] --label LABEL
                         (--version VERSION | --weights WEIGHTS)

options:
  -h, --help           show this help message and exit
  --workload WORKLOAD
  --project PROJECT
  --label LABEL
  --version VERSION
  --weights WEIGHTS

===== somm prompt promote =====
usage: somm prompt promote [-h] --workload WORKLOAD [--project PROJECT] --version VERSION --to TO
                           [--min-graded MIN_GRADED] [--min-score MIN_SCORE] [--force]

options:
  -h, --help            show this help message and exit
  --workload WORKLOAD
  --project PROJECT
  --version VERSION
  --to TO
  --min-graded MIN_GRADED
  --min-score MIN_SCORE
  --force

===== somm prompt score =====
usage: somm prompt score [-h] --workload WORKLOAD [--project PROJECT] [--version VERSION |
                         --label LABEL]

options:
  -h, --help           show this help message and exit
  --workload WORKLOAD
  --project PROJECT
  --version VERSION
  --label LABEL

===== somm eval =====
usage: somm eval [-h] [--project PROJECT] {promote-call,import,run} ...

positional arguments:
  {promote-call,import,run}
    promote-call        copy a sampled call into a durable eval dataset
    import              import reviewed JSONL prompt/expected-response pairs
    run                 run a workload against a durable eval dataset

options:
  -h, --help            show this help message and exit
  --project PROJECT

===== somm eval promote-call =====
usage: somm eval promote-call [-h] [--project PROJECT] --dataset DATASET
                              [--description DESCRIPTION]
                              call_id

positional arguments:
  call_id

options:
  -h, --help            show this help message and exit
  --project PROJECT
  --dataset DATASET
  --description DESCRIPTION

===== somm eval import =====
usage: somm eval import [-h] --workload WORKLOAD --dataset DATASET --file FILE [--project PROJECT]
                        [--description DESCRIPTION] [--json]

options:
  -h, --help            show this help message and exit
  --workload WORKLOAD
  --dataset DATASET
  --file FILE
  --project PROJECT
  --description DESCRIPTION
  --json

===== somm eval run =====
usage: somm eval run [-h] --workload WORKLOAD --dataset DATASET [--project PROJECT]
                     [--threshold THRESHOLD] [--max-tokens MAX_TOKENS] [--temperature TEMPERATURE]
                     [--provider PROVIDER] [--model MODEL]
                     [--reasoning-effort {none,minimal,low,medium,high,xhigh,max}]
                     [--implementation IMPLEMENTATION] [--judge-config JUDGE_CONFIG] [--json]

options:
  -h, --help            show this help message and exit
  --workload WORKLOAD
  --dataset DATASET
  --project PROJECT
  --threshold THRESHOLD
  --max-tokens MAX_TOKENS
  --temperature TEMPERATURE
  --provider PROVIDER
  --model MODEL
  --reasoning-effort {none,minimal,low,medium,high,xhigh,max}
                        reasoning budget for models that dial it per request; sweeping this across
                        runs is how a model gets evaluated rather than assumed
  --implementation IMPLEMENTATION
                        explicit implementation coordinate, normally a Git commit SHA
  --judge-config JUDGE_CONFIG
                        JSON binary-rubric criteria and explicit judge panel
  --json

===== somm optimize =====
usage: somm optimize [-h] --workload WORKLOAD [--project PROJECT] [--from FROM_REF]
                     [--threshold THRESHOLD] [--limit LIMIT] [--label LABEL]
                     [--max-tokens MAX_TOKENS] [--temperature TEMPERATURE] [--provider PROVIDER]
                     [--model MODEL]

options:
  -h, --help            show this help message and exit
  --workload WORKLOAD
  --project PROJECT
  --from FROM_REF
  --threshold THRESHOLD
  --limit LIMIT
  --label LABEL
  --max-tokens MAX_TOKENS
  --temperature TEMPERATURE
  --provider PROVIDER
  --model MODEL

===== somm campaign =====
usage: somm campaign [-h] [--project PROJECT] {run} ...

positional arguments:
  {run}
    run              run repeated eval rounds with keep/revert logging

options:
  -h, --help         show this help message and exit
  --project PROJECT

===== somm campaign run =====
usage: somm campaign run [-h] --workload WORKLOAD --dataset DATASET [--project PROJECT]
                         [--name NAME] [--metric {mean_score,pass_rate,error_rate}]
                         [--direction {gte,lte}] [--threshold THRESHOLD]
                         [--eval-threshold EVAL_THRESHOLD] [--max-rounds MAX_ROUNDS]
                         [--token-budget TOKEN_BUDGET] [--plateau-window PLATEAU_WINDOW]
                         [--min-delta MIN_DELTA] [--max-tokens MAX_TOKENS]
                         [--temperature TEMPERATURE] [--provider PROVIDER] [--model MODEL]
                         [--log LOG] [--json]

options:
  -h, --help            show this help message and exit
  --workload WORKLOAD
  --dataset DATASET
  --project PROJECT
  --name NAME
  --metric {mean_score,pass_rate,error_rate}
  --direction {gte,lte}
  --threshold THRESHOLD
  --eval-threshold EVAL_THRESHOLD
  --max-rounds MAX_ROUNDS
  --token-budget TOKEN_BUDGET
  --plateau-window PLATEAU_WINDOW
  --min-delta MIN_DELTA
  --max-tokens MAX_TOKENS
  --temperature TEMPERATURE
  --provider PROVIDER
  --model MODEL
  --log LOG
  --json

===== somm inbox =====
usage: somm inbox [-h] [--project PROJECT] [--workload WORKLOAD] [--all] [--json]
                  {list,apply,dismiss,show} ...

positional arguments:
  {list,apply,dismiss,show}
    list                list recommendations
    apply               apply a recommendation
    dismiss             dismiss one or more recommendations
    show                print one recommendation with its evidence

options:
  -h, --help            show this help message and exit
  --project PROJECT
  --workload WORKLOAD
  --all
  --json

===== somm inbox list =====
usage: somm inbox list [-h] [--project PROJECT] [--workload WORKLOAD] [--all] [--json]

options:
  -h, --help           show this help message and exit
  --project PROJECT
  --workload WORKLOAD
  --all
  --json

===== somm inbox apply =====
usage: somm inbox apply [-h] [--project PROJECT] [--json] recommendation_id

positional arguments:
  recommendation_id

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --json

===== somm inbox dismiss =====
usage: somm inbox dismiss [-h] [--project PROJECT] [--json]
                          recommendation_id [recommendation_id ...]

positional arguments:
  recommendation_id

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --json

===== somm inbox show =====
usage: somm inbox show [-h] [--project PROJECT] [--json] recommendation_id

positional arguments:
  recommendation_id

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --json

===== somm plugin =====
usage: somm plugin [-h] {list,info} ...

positional arguments:
  {list,info}
    list       list reference plugins, active hooks, and providers
    info       show one reference plugin

options:
  -h, --help   show this help message and exit

===== somm plugin list =====
usage: somm plugin list [-h]

options:
  -h, --help  show this help message and exit

===== somm plugin info =====
usage: somm plugin info [-h] name

positional arguments:
  name        reference plugin name

options:
  -h, --help  show this help message and exit

===== somm mcp =====
usage: somm mcp [-h] {install,uninstall,status} ...

positional arguments:
  {install,uninstall,status}
    install             install the somm MCP server
    uninstall           remove the somm MCP server
    status              show whether the somm MCP server is set up

options:
  -h, --help            show this help message and exit

===== somm mcp install =====
usage: somm mcp install [-h] [--harness {claude,codex}] [--print] [--scope {user,project}]

options:
  -h, --help            show this help message and exit
  --harness {claude,codex}
                        default: every supported harness found on PATH
  --print               show what would be done; change nothing
  --scope {user,project}
                        claude only (codex is user-level); default: user

===== somm mcp uninstall =====
usage: somm mcp uninstall [-h] [--harness {claude,codex}] [--print] [--scope {user,project}]

options:
  -h, --help            show this help message and exit
  --harness {claude,codex}
                        default: every supported harness found on PATH
  --print               show what would be done; change nothing
  --scope {user,project}
                        claude only (codex is user-level); default: user

===== somm mcp status =====
usage: somm mcp status [-h] [--harness {claude,codex}]

options:
  -h, --help            show this help message and exit
  --harness {claude,codex}
                        default: every supported harness found on PATH

===== somm skill =====
usage: somm skill [-h] {install,uninstall,status} ...

positional arguments:
  {install,uninstall,status}
    install             install the somm skill
    uninstall           remove the somm skill
    status              show whether the somm skill is set up

options:
  -h, --help            show this help message and exit

===== somm skill install =====
usage: somm skill install [-h] [--harness {claude,codex}] [--print]

options:
  -h, --help            show this help message and exit
  --harness {claude,codex}
                        default: every supported harness found on PATH
  --print               show what would be done; change nothing

===== somm skill uninstall =====
usage: somm skill uninstall [-h] [--harness {claude,codex}] [--print]

options:
  -h, --help            show this help message and exit
  --harness {claude,codex}
                        default: every supported harness found on PATH
  --print               show what would be done; change nothing

===== somm skill status =====
usage: somm skill status [-h] [--harness {claude,codex}]

options:
  -h, --help            show this help message and exit
  --harness {claude,codex}
                        default: every supported harness found on PATH

===== somm doctor =====
usage: somm doctor [-h] [--project PROJECT] [--max-exhausted-rate FRACTION] [--global]

options:
  -h, --help            show this help message and exit
  --project PROJECT
  --max-exhausted-rate FRACTION
                        regression gate: exit nonzero when the trailing-24h SommProvidersExhausted
                        rate exceeds FRACTION (e.g. 0.30). The rate is always printed; this flag
                        only affects the exit code — suitable for a cron/timer health check.
  --global              check ~/.somm/global.sqlite (every mirroring project) instead of this
                        project

===== somm loop =====
usage: somm loop [-h] [--project PROJECT] {run,install,uninstall,status} ...

positional arguments:
  {run,install,uninstall,status}
    run                 one pass: run the workers that are due, then exit
    install             write a systemd user timer that runs `somm loop run`
    uninstall           disable the timer for this project
    status              timer state and worker heartbeats

options:
  -h, --help            show this help message and exit
  --project PROJECT

===== somm loop run =====
usage: somm loop run [-h] [--project PROJECT] [--force] [--only {model_intel,shadow_eval,agent}]
                     [--all-projects] [--json]

options:
  -h, --help            show this help message and exit
  --project PROJECT
  --force               run every worker regardless of staleness
  --only {model_intel,shadow_eval,agent}
                        restrict to one worker (repeatable)
  --all-projects        every registered project ledger, not just this one
  --json

===== somm loop install =====
usage: somm loop install [-h] [--project PROJECT] [--every EVERY] [--no-enable]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --every EVERY      timer cadence (systemd span, default 15min)
  --no-enable        write units without enabling

===== somm loop uninstall =====
usage: somm loop uninstall [-h] [--project PROJECT]

options:
  -h, --help         show this help message and exit
  --project PROJECT

===== somm loop status =====
usage: somm loop status [-h] [--project PROJECT] [--json]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --json

===== somm prune =====
usage: somm prune [-h] [--project PROJECT] --older-than DAYS [--dry-run] [--vacuum] [--global]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --older-than DAYS  retention window in days
  --dry-run          report counts without deleting
  --vacuum           reclaim file space afterwards
  --global           prune ~/.somm/global.sqlite instead of this project's ledger

===== somm ledgers =====
usage: somm ledgers [-h] [--scan DIR] [--json] {forget,adopt} ...

positional arguments:
  {forget,adopt}
    forget        remove a registration (never deletes a file)
    adopt         register an existing ledger

options:
  -h, --help      show this help message and exit
  --scan DIR      also list calls.sqlite files under DIR that are not registered
  --json

===== somm ledgers forget =====
usage: somm ledgers forget [-h] [--path PATH] [--dry-run] PROJECT

positional arguments:
  PROJECT

options:
  -h, --help   show this help message and exit
  --path PATH  forget only this one of its paths
  --dry-run

===== somm ledgers adopt =====
usage: somm ledgers adopt [-h] [--project NAME] [--dry-run] PATH

positional arguments:
  PATH

options:
  -h, --help      show this help message and exit
  --project NAME  register under this name (default: every project the file holds)
  --dry-run

===== somm fleet =====
usage: somm fleet [-h] {status} ...

positional arguments:
  {status}
    status    versions, schema skew, loop, unknown cost (as in doctor --global)

options:
  -h, --help  show this help message and exit

===== somm fleet status =====
usage: somm fleet status [-h] [--json]

options:
  -h, --help  show this help message and exit
  --json

===== somm mirror =====
usage: somm mirror [-h] [--project PROJECT] {sync} ...

positional arguments:
  {sync}
    sync             copy call rows a registered ledger has and the mirror lacks

options:
  -h, --help         show this help message and exit
  --project PROJECT  only this project's ledgers and rows

===== somm mirror sync =====
usage: somm mirror sync [-h] [--project PROJECT] [--dry-run] [--batch BATCH]

options:
  -h, --help         show this help message and exit
  --project PROJECT  only this project's ledgers and rows
  --dry-run          count, copy nothing
  --batch BATCH      rows per transaction

===== somm alias =====
usage: somm alias [-h] [--project PROJECT] [--global] {list,seed,set} ...

positional arguments:
  {list,seed,set}
    list             show alias rows
    seed             derive aliases by rule from routes seen in the ledger
    set              map one provider/model route to a canonical id

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --global           operate on ~/.somm/global.sqlite

===== somm alias list =====
usage: somm alias list [-h] [--json] [--project PROJECT] [--global]

options:
  -h, --help         show this help message and exit
  --json
  --project PROJECT
  --global           operate on ~/.somm/global.sqlite

===== somm alias seed =====
usage: somm alias seed [-h] [--project PROJECT] [--global]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --global           operate on ~/.somm/global.sqlite

===== somm alias set =====
usage: somm alias set [-h] [--project PROJECT] [--global] provider model canonical

positional arguments:
  provider
  model
  canonical

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --global           operate on ~/.somm/global.sqlite

===== somm serve =====
usage: somm serve [-h] [--project PROJECT] [--host HOST] [--port PORT]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --host HOST
  --port PORT

===== somm spend =====
usage: somm spend [-h] [--project PROJECT] [--json]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --json             emit a JSON array instead of the aligned table

===== somm climb =====
usage: somm climb [-h] [--workload WORKLOAD] [--since SINCE] [--project PROJECT] [--json] [path]

positional arguments:
  path                 receipts directory or file (default: $SOMM_CLIMB_RECEIPTS)

options:
  -h, --help           show this help message and exit
  --workload WORKLOAD  only this socket id; also the exact somm workload name to join by
  --since SINCE        window in days (default 30)
  --project PROJECT
  --json

===== somm gpu =====
usage: somm gpu [-h] [--project PROJECT] {spend} ...

positional arguments:
  {spend}
    spend            GPU spend by day and workload; failed jobs; ledger match

options:
  -h, --help         show this help message and exit
  --project PROJECT

===== somm gpu spend =====
usage: somm gpu spend [-h] [--project PROJECT] [--since SINCE] [--path PATH] [--json]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --since SINCE      window in days (default 30)
  --path PATH        receipts directory (default: $SOMM_GIPPER_RECEIPTS, else
                     $GIPPER_DATA/receipts, else ~/.local/share/gipper/receipts)
  --json

===== somm backfill-costs =====
usage: somm backfill-costs [-h] [--project PROJECT] [--since SINCE] [--dry-run]

options:
  -h, --help         show this help message and exit
  --project PROJECT
  --since SINCE      only calls from the last N days
  --dry-run          report without writing

===== somm plans =====
usage: somm plans [-h] [--project PROJECT] [--project-only] [--json] [--catalog] [--learn]
                  [--dry-run] [--min-events MIN_EVENTS]

options:
  -h, --help            show this help message and exit
  --project PROJECT
  --project-only        count only this project's usage (default: whole fleet — quotas are shared)
  --json
  --catalog             list known plans from the bundled catalog (with sources + verified dates)
  --learn               write observed quota ceilings from recent 429s into plans.toml
  --dry-run             show learned limits without writing
  --min-events MIN_EVENTS
                        minimum quota-error events required to learn a ceiling

===== somm drain-spool =====
usage: somm drain-spool [-h] [--project PROJECT]

options:
  -h, --help         show this help message and exit
  --project PROJECT
