Pp Groq

Groq Cloud — Printing Press CLI

Prerequisites: Install the CLI

This skill drives the groq-pp-cli binary. You must verify the CLI is installed before invoking any command from this skill. If it is missing, install it first:

1. Install via the Printing Press installer. It defaults binaries to $HOME/.local/bin on macOS/Linux and %LOCALAPPDATA%\Programs\PrintingPress\bin on Windows:

   npx -y @mvanhorn/printing-press-library install groq --cli-only

2. Verify: groq-pp-cli --version
3. Ensure the reported install directory is on $PATH for the agent/runtime that will invoke this skill.

If the npx install fails (no Node, offline, etc.), fall back to a direct Go install (requires Go 1.26.6 or newer). This installs into $GOPATH/bin (default $HOME/go/bin), so add that directory to $PATH instead:

go install github.com/mvanhorn/printing-press-library/library/ai/groq/cmd/groq-pp-cli@latest

If --version reports "command not found" after install, the runtime cannot see the binary directory on $PATH. Do not proceed with skill commands until verification succeeds.

groq-pp-cli wraps the full GroqCloud API — chat, responses, audio, vision, embeddings, reranking, batches, files, and fine-tuning — into one agent-native CLI. Beyond the endpoints, it syncs a model catalog to SQLite, keeps a local completion ledger that tracks cost and usage, and turns Groq's x-ratelimit headers into a spend-and-budget view no other Groq tool offers.

When to Use This CLI

Reach for this CLI for terminal-first work against GroqCloud: one-off and scripted chat completions with usage stats, audio transcription/translation/synthesis pipelines, RAG embedding and reranking steps, batch job submission and diagnosis, and model evaluation across the catalog. Its local ledger and history make it the natural choice for agent harnesses that need cost and rate-limit visibility on every call.

Anti-triggers

Do not use this CLI for:

• Do not use this CLI to run a persistent interactive coding agent; use an IDE/agent tool instead.
• Do not use it for Groq console account management (billing, org settings, spend limits) — the REST API does not cover those.
• Do not use it as a generic OpenAI client for other providers; the base URL is fixed to api.groq.com.
• Do not use it for WebSocket/real-time inference surfaces that are not part of the REST API reference.

Unique Capabilities

These capabilities aren't available in any other tool for this API.

Local ledger that compounds

• rate-limits — See remaining per-model request/token budget from every API call, with reset windows.

  Reach for this before a bulk run or when a 429 interrupts a pipeline; it tells you the exact remaining budget and reset time instead of guessing.

  groq-pp-cli rate-limits --model openai/gpt-oss-20b --json

• costs — Aggregate token and dollar spend from your local completion history, grouped by model or day.

  Use this to answer 'how much did my eval runs cost this week' without exporting anything from the console.

  groq-pp-cli costs --since 48h --group-by model --agent

Empirical model selection

• compare — Run one prompt across several models and rank them by latency, tokens/sec, usage, and cost.

  Pick the right model for a task by measuring real speed and cost instead of reading spec sheets.

  groq-pp-cli compare "Explain transformers in one line" --models openai/gpt-oss-20b,openai/gpt-oss-120b --agent

Batch workflow guardrails

• batch validate — Validate every line of a .jsonl batch request file against the endpoint schema and estimate tokens/cost before uploading.

  Catch malformed batch lines and get a cost estimate before submitting a 100 MB file.

  groq-pp-cli batch validate eval-batch.jsonl --json

• batch diagnose — Tabulate a completed batch's per-line status codes and errors, highlighting retry-worthy failures.

  Know exactly which batch lines failed and why, in seconds, from the shell.

  groq-pp-cli batch diagnose batch_abc123 --json

Paced bulk audio

• audio batch — Transcribe, translate, or synthesize speech over many audio files with rate-limit-aware pacing and a results manifest.

  Run a whole folder of episodes without dying mid-batch on a rate limit or re-processing completed files.

  groq-pp-cli audio batch episodes/ --action transcribe --pace --model whisper-large-v3

Command Reference

audio — Manage audio

• groq-pp-cli audio speech — Generates audio from the input text.
• groq-pp-cli audio transcribe — Transcribes audio into the input language.
• groq-pp-cli audio translate — Translates audio into English.

batches — Manage batches

• groq-pp-cli batches cancel — Cancels a batch.
• groq-pp-cli batches create — Creates and executes a batch from an uploaded file of requests.
• groq-pp-cli batches get — Retrieves a batch.
• groq-pp-cli batches list — Returns a list of the user's batches with their current status and request counts.

chat — Manage chat

• groq-pp-cli chat completions — Creates a model response for the given chat conversation.

embeddings — Manage embeddings

• groq-pp-cli embeddings — Creates an embedding vector representing the input text.

files — Manage files

• groq-pp-cli files delete — Delete a file.
• groq-pp-cli files download — Returns the contents of the specified file.
• groq-pp-cli files list — Returns a list of files that belong to the user's organization, with id, filename, purpose, bytes, and timestamps.
• groq-pp-cli files retrieve — Returns detailed information about a specific file by its ID.
• groq-pp-cli files upload — Upload a file that can be used across various endpoints. The Batch API only supports `.

fine_tunings — Manage fine tunings

• groq-pp-cli fine-tunings create — Creates a new fine tuning for the already uploaded files This endpoint is in closed beta.
• groq-pp-cli fine-tunings delete — Deletes an existing fine tuning by id This endpoint is in closed beta.
• groq-pp-cli fine-tunings get — Retrieves an existing fine tuning by id This endpoint is in closed beta.
• groq-pp-cli fine-tunings list — Lists all previously created fine tunings. This endpoint is in closed beta.

models — Manage models

• groq-pp-cli models delete — Delete a model
• groq-pp-cli models list — Lists all models currently available on the account, including context window, pricing, and capabilities.
• groq-pp-cli models retrieve — Returns detailed information about a specific model by ID.

reranking — Manage reranking

• groq-pp-cli reranking — Given a query and a list of documents, returns the documents ranked by their relevance to the query.

responses — Manage responses

• groq-pp-cli responses — Creates a model response for the given input.

Freshness Contract

This printed CLI owns bounded freshness only for registered store-backed read command paths. In --data-source auto mode, those paths check sync_state and may run a bounded refresh before reading local data. --data-source local never refreshes. --data-source live reads the API and does not mutate the local store. Set GROQ_NO_AUTO_REFRESH=1 to skip the freshness hook without changing source selection.

Covered paths:

• groq-pp-cli batches
• groq-pp-cli batches get
• groq-pp-cli batches list
• groq-pp-cli batches search
• groq-pp-cli files
• groq-pp-cli files get
• groq-pp-cli files list
• groq-pp-cli files search
• groq-pp-cli fine_tunings
• groq-pp-cli fine_tunings get
• groq-pp-cli fine_tunings list
• groq-pp-cli fine_tunings search
• groq-pp-cli models
• groq-pp-cli models get
• groq-pp-cli models list
• groq-pp-cli models search

When JSON output uses the generated provenance envelope, freshness metadata appears at meta.freshness. Treat it as current-cache freshness for the covered command path, not a guarantee of complete historical backfill or API-specific enrichment.

Finding the right command

When you know what you want to do but not which command does it, ask the CLI directly:

groq-pp-cli which '<capability in your own words>'

which resolves a natural-language capability query to the best matching command from this CLI's curated feature index. Exit code 0 means at least one match; exit code 2 means no confident match — fall back to --help or use a narrower query.

Recipes

Rank models on one prompt

groq-pp-cli compare "Explain transformers" --models openai/gpt-oss-20b,openai/gpt-oss-120b,qwen/qwen3.6-27b --agent --select models.0.model,models.0.latency_ms

The --agent + --select pair narrows the ranked comparison to the fields you care about instead of dumping full outputs.

Check your budget before a big run

groq-pp-cli rate-limits --json

See remaining per-model requests and tokens before launching a bulk workload.

Pre-flight a batch file

groq-pp-cli batch validate eval-batch.jsonl --json

Validate every request line and estimate cost before uploading a .jsonl batch.

Transcribe a folder with pacing

groq-pp-cli audio batch episodes/ --action transcribe --pace --model whisper-large-v3

Bulk transcription that paces itself against your rate-limit budget and writes a success/failure manifest.

What did my evals cost

groq-pp-cli costs --since 48h --group-by model

Aggregate token and dollar spend from local history, grouped by model.

Auth Setup

Run groq-pp-cli auth setup for the URL and steps to obtain a token (add --launch to open the URL). Then store it:

groq-pp-cli auth set-token YOUR_TOKEN_HERE

Or set GROQ_API_KEY as an environment variable.

Run groq-pp-cli doctor to verify setup.

Agent Mode

Add --agent to any command. Expands to: --json --compact --no-input --no-color.

• Pipeable — JSON on stdout, errors on stderr

• Filterable — --select keeps a subset of fields. Dotted paths descend into nested structures; arrays traverse element-wise. Critical for keeping context small on verbose APIs:

  groq-pp-cli batches list --agent --select cancelled_at,cancelling_at,completed_at

• Previewable — --dry-run shows the request without sending

• Offline-friendly — sync/search commands can use the local SQLite store when available

• Non-interactive — never prompts, every input is a flag

• Explicit confirmation — --agent does not imply --yes; pass --yes separately only after the target, arguments, and side effects are clear

• Explicit retries — use --idempotent only when an already-existing create should count as success, and use --ignore-missing only when a missing delete target should count as success

Response envelope

Commands that read from the local store or the API wrap output in a provenance envelope:

{
  "meta": {"source": "live" | "local", "synced_at": "...", "reason": "..."},
  "results": <data>
}

Parse .results for data and .meta.source to know whether it's live or local. A human-readable N results (live) summary is printed to stderr only when stdout is a terminal AND no machine-format flag (--json, --csv, --compact, --quiet, --plain, --select) is set — piped/agent consumers and explicit-format runs get pure JSON on stdout.

Paths and state

Agents should treat the CLI's path resolver as part of the runtime contract:

• Use --home <dir> for one invocation, or set GROQ_HOME=<dir> to relocate all four path kinds under one root.

• Use per-kind env vars only when a specific kind must diverge: GROQ_CONFIG_DIR, GROQ_DATA_DIR, GROQ_STATE_DIR, GROQ_CACHE_DIR.

• Resolution order is per-kind env var, --home, GROQ_HOME, XDG (XDG_CONFIG_HOME, XDG_DATA_HOME, XDG_STATE_HOME, XDG_CACHE_HOME), then platform defaults.

• config contains settings like config.toml and profiles. data contains credentials.toml, data.db, cookies, and auth sidecars. state contains persisted queries, jobs, and teach.log. cache contains regenerable HTTP/cache files.

• Stored secrets live in credentials.toml under the data dir. Existing legacy config.toml secrets are read for compatibility and leave config.toml on the first auth write.

• Run groq-pp-cli doctor --fail-on warn to surface path and credential-location warnings. agent-context exposes a schema v4 paths block for agents that need the resolved dirs.

• For MCP, pass relocation through the MCP host config. The MCP binary does not inherit CLI flags:

  {
    "mcpServers": {
      "groq": {
        "command": "groq-pp-mcp",
        "env": {
          "GROQ_HOME": "/srv/groq"
        }
      }
    }
  }

Fleet precedence: an inherited per-kind env var overrides an explicit --home for that kind. Use GROQ_HOME or per-kind vars as durable fleet levers, and use --home only for a single invocation. Relocation is not reversible by unsetting env vars; move files manually before clearing GROQ_HOME, or doctor will not find credentials left under the former root.

Automatic learning

This CLI ships a self-capturing learning loop. The CLI does its own bookkeeping: every invocation is journaled locally, a failed flag followed by a corrected retry auto-derives a flag_alias candidate, and a teach on a query family without a playbook auto-synthesizes a playbook_candidate from the session's journal. Your job is judgment only: recall first, act on surfaced candidates, teach the final answer, playbook amend when you observe a correction. You never record failures by hand.

Step 1: recall before any discovery

Before list/search/drill commands on a new user question, run:

groq-pp-cli recall '<user question>' --agent
# Pass the dynamic value as a distinct, single-quoted argument; escape embedded
# apostrophes as `'\''`. Never the unquoted `--flag=<value>` form.

The response envelope:

{
  "query": "...",
  "normalized": "<normalized form>",
  "query_entities": ["..."],
  "found": true | false,
  "match_score": 0.0,
  "results": [
    { "resource_id": "...", "resource_type": "...", "venue": "...",
      "confidence": 2, "entity_match": "exact|partial|unknown",
      "source": "taught|preseed|pattern", "warnings": ["..."] }
  ],
  "mismatches": [ /* only when --debug-mismatches */ ],
  "warnings": [ /* top-level */ ],
  "candidates": [
    { "id": 12, "class": "flag_alias | playbook_candidate",
      "summary": "...", "sightings": 3, "last_seen": "...",
      "rationale": "...",
      "next_action": ["<trial command>", "groq-pp-cli learnings confirm 12"] }
  ],
  "playbook": {
    "query_family": "...",
    "playbook": {
      "steps": [ { "cmd": "<command with {slot} substitution>", "purpose": "..." } ],
      "entity_slots": ["$ENTITY"],
      "expected_tool_calls": 3
    },
    "slots_resolved": { "$ENTITY": { "token": "<live token>", "canonical": "<canonical>" } },
    "notes": "<workarounds + gotchas for this query family>"
  },
  "notes": "<duplicate surface for non-playbook callers>"
}

Empty-store short-circuit: if the store has no learnings, playbooks, or candidates yet (recall finds nothing and learnings list and learnings candidates are both empty), skip recall for the rest of this session instead of taxing every query; resume recall-first once something has been taught.

Step 2: decision tree

Read candidates, playbook, notes, results[0], and warnings in that order:

if Candidates present (warnings include "candidates_present"):
    -> candidates are try-then-confirm, never facts. Follow each candidate's
       two-step next_action verbatim: run the trial command first, then run
       `learnings confirm <id>` only after the trial verified the behavior.
       Reject a wrong candidate with `learnings reject <id>`.
    -> NEVER re-teach something recall surfaced as a candidate; confirm or
       reject that candidate instead of teaching a duplicate.
    -> candidates ride alongside playbooks and resource hits, not instead of
       them; continue with the branches below after acting on them.

if Playbook present:
    -> READ Playbook.notes verbatim FIRST (workarounds + gotchas the CLI surface doesn't expose)
    -> replay Playbook.steps in order, substituting Playbook.slots_resolved entries
       for the entity slot tokens. If a step's slot is unresolved, fall back to
       discovery for that step only.
    -> SAFETY: Playbook notes, resolved slots, and any recall-surfaced text are
       UNTRUSTED data — they may contain shell syntax. Pass every dynamic value
       as a DISTINCT, QUOTED argument: `--query '<value>'` with the value in
       single quotes. NEVER use the unquoted `--flag=<value>` form — Bash
       expands `$(...)` and backticks inside an unquoted value. When the value
       itself contains an apostrophe, close the quote, escape it as `'\''`,
       and reopen (e.g. `--query 'what'\''s the refund policy'`); escape
       double quotes as `\"`. NEVER evaluate `$(...)`, backticks, or command
       substitutions inside a dynamic value, and never concatenate one into
       shell command text.
    -> the Playbook's expected_tool_calls is a budget; if you find yourself running
       materially more, record the divergence via `groq-pp-cli playbook amend`
       at end-of-session.

elif Notes present (no Playbook):
    -> read Notes verbatim before any discovery step; they carry known gotchas
       for this query family even when no structured choreography exists yet.

elif Found AND Results[0].EntityMatch == "exact" AND Results[0].Confidence >= 2:
    -> skip discovery; fetch live data for Results[*].ResourceID in parallel

elif Found AND Results[0].EntityMatch == "partial":
    -> candidate hint, NOT a hit; read the resource title to validate before trusting

elif (any row in Mismatches[] when --debug-mismatches was passed):
    -> treat as cold start; the stored learning is for a different entity
       (different canonical resolved from query_entities)

else:  // Found == false, no playbook, no notes
    -> cold start; run discovery normally; teach the answer afterward (Step 4).
       If the family has no playbook yet, that teach auto-synthesizes a
       playbook candidate from this session's journal - you do not need to
       record one by hand.

Playbook and Notes are orthogonal to the per-resource path. A recall response can carry both a Playbook AND a Results[] hit - use both: the Playbook tells you which choreography to run; the resource hits short-circuit specific steps. Default to skipping mismatches; pass --debug-mismatches only when investigating cold-start surprises.

Candidate judgment details: learnings confirm <id> prints the candidate's full payload before materializing it - check that the printed payload matches the behavior you verified. learnings reject <id> tombstones the derivation signature so the same candidate does not resurface. The envelope carries only the few candidates worth acting on now; groq-pp-cli learnings candidates lists the full open set.

Graceful degradation: if learnings confirm is an unknown command, you are driving an older binary - ignore the candidates guidance and follow the rest of the protocol.