Pp Sarvam

Sarvam AI — Printing Press CLI

Prerequisites: Install the CLI

This skill drives the sarvam-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 sarvam --cli-only

2. Verify: sarvam-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 before this CLI has a public-library category, install Node or use the category-specific Go fallback after publish.

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.

Sarvam AI's official SDKs and MCP server are great for code, but nothing offers offline capability: no local history of translations, TTS generations, transcriptions, or chat threads. sarvam-pp-cli adds a local SQLite store, voice auditioning, conversation resume, batch job retry/report, pronunciation spot-checks, subtitle export, and a doc-ai extraction schema library — all with --json, --dry-run, and typed exit codes for agents and scripts.

When to Use This CLI

Use sarvam-pp-cli when you need to translate, transliterate, transcribe, synthesize speech, chat, or extract document fields through Sarvam AI's Indic-language models — especially when you want a record of what you did, offline search across past work, batch orchestration, or scriptable/agent-friendly output. It is the right tool for voice-agent prompt engineering, call-center transcription QA, content localization, and document-intelligence pipelines.

Anti-triggers

Do not use this CLI for:

• Do not use this CLI for the Voice Agents platform (apps.sarvam.ai) — deployments, campaigns, cohorts, analytics, and instant outbound use a separate X-API-Key auth system with org/workspace scoping and are not implemented here.
• Do not use this CLI for real-time WebSocket streaming (realtime STT, TTS WS) — those are AsyncAPI surfaces not covered by the REST CLI.
• Do not use this CLI to replace a full speech-recognition model benchmark — the API is a paid service and the CLI does not add model-evaluation tooling.

Unique Capabilities

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

Speech workflow

• voices preview — Generate one sample sentence across every TTS speaker and hear them all side by side

  Use when choosing a TTS voice for a new language or prompt set without burning manual API calls

  sarvam-pp-cli voices preview --lang hi-IN --sample "नमस्ते, स्वागत है" --speakers shubh,ritu,priya --output ./voices

• pron-check — Verify a term's TTS pronunciation via a speech round-trip (TTS then STT)

  Use to confirm a pronunciation dictionary edit took effect before shipping IVR prompts

  sarvam-pp-cli pron-check "SarvamPay" --lang hi-IN

Local state that compounds

• chat resume — Continue a past chat thread from local history with full context

  Use to continue an assistant session without losing context. New chats save the request messages and reply in local SQLite; older response-only records can resume from the last reply but cannot recover earlier prompts. This command sends a new paid chat request.

  sarvam-pp-cli chat resume 20260814_2d09e061 "what was our conclusion?"

  Completed text-only streamed chats can also be resumed. JSON output includes structured results.id, results.choices, and results.usage, plus every SSE event in results.stream; only the first text choice is saved for resume. Incomplete streams and streamed tool calls are not saved because their full context cannot be reconstructed safely. Successful text-to-speech requests save their request text without copying audio into SQLite.

• subs — Emit .srt/.vtt subtitles from timestamped transcriptions in local history

  Use to turn a timestamped transcription into subtitles without a throwaway script

  sarvam-pp-cli subs --from last --format srt --output subtitles.srt

Job orchestration

• stt-job retry — Re-run only the failed files of a batch STT job with one command

  Use when a batch job partially fails and you need to reprocess just the failures

  sarvam-pp-cli stt-job retry 20260707_9f1c2b3a-4d5e-6f70-8a9b-c0d1e2f3a4b5 --failed-only --dir ./audio/

• stt-job report — Per-file digest of a batch STT job with typed exit codes for cron alerting

  Use in cron to alert when a batch transcription job degrades

  sarvam-pp-cli stt-job report 20260707_9f1c2b3a-4d5e-6f70-8a9b-c0d1e2f3a4b5 --json

Document intelligence

• docai schema list — Save, list, and diff doc-ai extraction schemas locally

  Use to version extraction schemas so schema changes never silently break extraction runs

  sarvam-pp-cli docai schema list

• docai batch — Run a saved extraction schema over a folder of documents with job pacing

  Use for weekly batch document extraction (KYC, invoices) without writing orchestration code

  sarvam-pp-cli docai batch --schema invoice-v1 --dir ./docs/ --out ./results/

Command Reference

chat — Manage chat

• sarvam-pp-cli chat — Creates a model response for the given chat conversation. Serves sarvam-105b and sarvam-105b-conversations models.

doc-ai — Manage doc ai

• sarvam-pp-cli doc-ai digitise — Creates and starts a digitisation job from files or pre-uploaded handles.
• sarvam-pp-cli doc-ai download-url — Returns a presigned URL to download the output of a completed doc-ai job.
• sarvam-pp-cli doc-ai extract — Creates and starts an extract job from files or pre-uploaded handles.
• sarvam-pp-cli doc-ai results — Fetches the results of a completed doc-ai job, including extracted fields and annotations with confidence scores.
• sarvam-pp-cli doc-ai status — Polls the status of a doc-ai job until a terminal status (completed, partially_completed, failed, rejected).
• sarvam-pp-cli doc-ai upload — Creates a presigned URL to upload a document for doc-ai processing.

models — Manage models

• sarvam-pp-cli models — Lists the model IDs this deployment currently serves.

speech-to-text — Manage speech to text

• sarvam-pp-cli speech-to-text speech-to-text — Transcribes speech to text in multiple Indian languages and English. Accepts an audio file via multipart form-data.
• sarvam-pp-cli speech-to-text stt-job-download — Returns presigned download URLs for the output files of a completed batch speech-to-text job.
• sarvam-pp-cli speech-to-text stt-job-initiate — Creates a new speech-to-text bulk job and returns a job UUID and storage details for processing multiple audio files.
• sarvam-pp-cli speech-to-text stt-job-start — Starts processing a speech-to-text bulk job after all audio files have been uploaded.
• sarvam-pp-cli speech-to-text stt-job-status — Returns the status of a batch speech-to-text job including per-file details and download information.
• sarvam-pp-cli speech-to-text stt-job-upload — Generates presigned upload URLs for audio files that will be processed in a speech-to-text bulk job.

text-lid — Manage text lid

• sarvam-pp-cli text-lid — Identifies the language (e.g. hi-IN) and script (e.g.

text-to-speech — Manage text to speech

• sarvam-pp-cli text-to-speech create-pronunciation-dictionary — Uploads a .json file to create a new pronunciation dictionary. Only supported by bulbul:v3.
• sarvam-pp-cli text-to-speech delete-pronunciation-dictionary — Deletes a pronunciation dictionary by ID.
• sarvam-pp-cli text-to-speech get-pronunciation-dictionary — Fetches a single pronunciation dictionary by ID.
• sarvam-pp-cli text-to-speech list-pronunciation-dictionaries — Lists all pronunciation dictionaries for the user. Dictionaries define custom word pronunciations used by bulbul:v3 TTS.
• sarvam-pp-cli text-to-speech stream — Converts the input text into a streamed spoken audio response using the specified output codec (e.g. MP3).
• sarvam-pp-cli text-to-speech text-to-speech — Converts text into spoken audio.
• sarvam-pp-cli text-to-speech update-pronunciation-dictionary — Updates an existing pronunciation dictionary with a new JSON file.

translate — Manage translate

• sarvam-pp-cli translate — Converts text from one language to another while preserving meaning. Supports 22 Indic languages plus English.

transliterate — Manage transliterate

• sarvam-pp-cli transliterate — Transliterates text from one script to another (e.g.

Finding the right command

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

sarvam-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

Translate a support reply

sarvam-pp-cli translate --input "Your EMI of Rs. 3000 is pending" --source-language-code en-IN --target-language-code hi-IN --mode formal

Translate a single message with formal tone for customer communications

Audition voices for a new prompt set

sarvam-pp-cli voices preview --lang hi-IN --sample "नमस्ते, स्वागत है" --speakers shubh,ritu,priya

Hear 3 voices on the same sample before committing to a speaker

Subtitle a promo video

sarvam-pp-cli subs --from last --format srt --output subtitles.srt

Turn the last timestamped transcription into SRT subtitles

Reprocess failed batch files

sarvam-pp-cli stt-job retry 20260707_9f1c2b3a-4d5e-6f70-8a9b-c0d1e2f3a4b5 --failed-only --dir ./audio/

Re-run only the files that failed in a batch transcription job

Check a pronunciation dictionary

sarvam-pp-cli pron-check "SarvamPay" --lang hi-IN --dict p_5cb7faa6

Verify a custom pronunciation actually changed how the term sounds

Extract fields from a folder of documents

sarvam-pp-cli docai batch --schema invoice-v1 --dir ./docs/ --out ./results/

Run a saved extraction schema over every document with pacing

Continue a past chat session

sarvam-pp-cli chat resume 20260814_2d09e061 "what was our conclusion?"

Pick up an assistant conversation where it left off, with full context

Auth Setup

Authentication uses the Sarvam AI API subscription key (sk_ format). Set it with export SARVAM_API_KEY=sk_... or sarvam-pp-cli auth set-token. The key goes in the api-subscription-key header (or Authorization: Bearer). Note: an invalid key returns HTTP 403 with invalid_api_key_error, not 401 — treat 403 as the auth-failure signal. The separate Voice Agents platform (apps.sarvam.ai) uses a different X-API-Key system and is out of scope for this CLI.

Run sarvam-pp-cli doctor to verify setup.

Agent Mode

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

• 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:

  sarvam-pp-cli models --agent --select created,id,object

• 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 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 SARVAM_HOME=<dir> to relocate all four path kinds under one root.

• Use per-kind env vars only when a specific kind must diverge: SARVAM_CONFIG_DIR, SARVAM_DATA_DIR, SARVAM_STATE_DIR, SARVAM_CACHE_DIR.

• Resolution order is per-kind env var, --home, SARVAM_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 sarvam-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": {
      "sarvam": {
        "command": "sarvam-pp-mcp",
        "env": {
          "SARVAM_HOME": "/srv/sarvam"
        }
      }
    }
  }

Fleet precedence: an inherited per-kind env var overrides an explicit --home for that kind. Use SARVAM_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 SARVAM_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:

sarvam-pp-cli recall "<user's question>" --agent

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>", "sarvam-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.
    -> the Playbook's expected_tool_calls is a budget; if you find yourself running
       materially more, record the divergence via `sarvam-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; sarvam-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.