Wavo troubleshooting

Last updated: May 19, 2026

Wavo troubleshooting

Common Wavo issues with diagnosis + actionable fix. Always check data freshness (ai-104) + credits balance (ai-109) first — these explain most surprises. For persistent bugs: give thumbs-down feedback + open a support ticket via Wavo.

Who is this for

Anyone hitting an unexpected Wavo behavior. First stop before contacting support.

Issue 1: Wavo gave a wrong metric

Symptoms

Wavo cited a ROAS / CPA / spend that doesn't match what the platform UI shows.

Causes + fixes

  1. Postback lag — purchase / ROAS / CPA data settles 24-72h after the conversion. Ask Wavo to "re-query for last 7d" instead of "today".

  2. Strategic context cache (5 min) — Wavo may have served slightly stale data. Ask again or specify date range to bust cache.

  3. Currency conversion — if Cross-Channel, values are in target_currency; platform UI shows native. Small variance expected.

  4. Wrong tool invoked — Wavo may have picked the wrong tool for ambiguous prompts. Be specific.

Fix: ask Wavo to cite the source tool + re-query with explicit date / platform / currency.

Issue 2: Stream interrupted mid-response

Symptoms

Wavo started streaming but stopped mid-sentence. Spinner stuck or error message.

Causes + fixes

  1. Network glitch — temporary disconnect; chat-resume.ts handles auto-reconnect for many cases

  2. Provider outage — Anthropic / OpenAI / etc. temporary issue; retry in 30s

  3. Token limit hit — very long output; ask Wavo to "continue" or shorten the question

Fix:

  • Click Resume button (if shown) — resumes from where stream stopped

  • OR send a follow-up: "continue" or "please finish"

  • OR re-send your original question

Issue 3: Approval card disappeared

Symptoms

Wavo presented an approval card; you waited to decide; now the card is gone.

Cause

ACTION_TTL_MS=1800000 (30 minutes). Approval cards expire after 30 min for safety.

Fix

Re-ask Wavo to do the action. A new approval card will be generated with fresh TTL.

If you intended to approve but missed: nothing was executed. No harm done.

See ai-105 actions.

Issue 4: "I don't have permission" error

Symptoms

Wavo says it can't perform an action you requested.

Causes + fixes

  1. Role-based accesschat-tool-rbac.ts enforces role hierarchy. Viewer can't trigger write tools, etc.

  2. Tool-specific permission — some tools require manager+ or owner+ role

  3. Workspace policy — admin may have restricted certain tools

Fix:

  • Check your role at /settings → Profile

  • Ask your admin to grant the needed role

  • For one-off: ask a teammate with higher role

Issue 5: Out of credits

Symptoms

Wavo returns 402 with reason insufficient_credits or monthly_cap_reached.

Causes + fixes

  1. Balance zero — total exhausted; admin tops up via Stripe

  2. Monthly cap reached — admin-set cap; raise via PUT /api/v1/chat/credits/cap or wait for cap_reset_at

  3. Failed payment — Stripe top-up may not have completed (3DS required)

Fix:

  • View at /settings/team/billing

  • Top up: POST /api/v1/chat/credits/checkout → Stripe

  • Raise cap: owner / admin via PUT /api/v1/chat/credits/cap

  • Confirm 3DS if Stripe required it: POST /api/v1/chat/credits/confirm

See ai-109 credits-limits.

Issue 6: Wavo doesn't use my uploaded file

Symptoms

Uploaded a file but Wavo doesn't reference it.

Causes + fixes

  1. Upload failed — check for error toast; verify under 10 MB + MIME-allowed

  2. MIME type not whitelisted — only image (jpeg/png/webp/gif), pdf, csv, plain text, docx, xlsx

  3. File too large — 10 MB max for chat upload (25 MB for voice)

  4. Wavo didn't see it as relevant — explicitly say "use the attached file"

Allowed MIME (POST /api/v1/chat-upload):

  • image/jpeg, image/png, image/webp, image/gif

  • application/pdf

  • text/csv, text/plain

  • docx (officedocument.wordprocessingml.document)

  • xlsx (officedocument.spreadsheetml.sheet)

Fix: re-upload smaller / supported format + explicitly reference the file in your message.

Issue 7: Voice transcription wrong language

Symptoms

Spoke Italian; Whisper returned text in another language. Or vice versa.

Cause

Whisper auto-detects but can fail with mixed-language audio. Default language is it (Italian) in chat-voice.route.ts.

Fix

Pass language parameter explicitly (BCP-47 code: en, es, fr, de, etc.) via the voice settings before recording.

Or: speak clearly in a single language; mixed-language audio is harder to transcribe.

Issue 8: Wavo hallucinated data

Symptoms

Wavo cited a campaign / metric / fact that doesn't exist.

Causes + fixes

  1. Vague prompt — Wavo tried to satisfy + extrapolated

  2. Model limitation — some models hallucinate more than others; switch to Sonnet / Opus / Gemini Pro

  3. Missing tool result — Wavo should have invoked a tool but didn't (rare; usually means tool failed silently)

Fix:

  • Ask Wavo to cite the source tool: "where did you get that number?"

  • If invented: "you made that up — please query the real data"

  • Give thumbs-down feedback + comment "hallucination"

  • Switch model in /settings → AI preferences

For persistent hallucinations: open support ticket via Wavo.

Issue 9: Wavo slow to respond

Symptoms

Response takes much longer than usual.

Causes + fixes

  1. Heavy model + complex query — Opus + 5 tool rounds = 30+ sec normal

  2. Provider load — peak hours; switch to faster model (Haiku / Flash / GPT-4.1 mini)

  3. Long context — uploaded files / long history = more processing

  4. Tool latency — tool may be slow (especially web_search or heavy DB queries)

Fix:

  • Switch model: /settings → AI preferences → preferred_model

  • Trim file uploads to essentials

  • Break complex queries into shorter sequential ones

Issue 10: Lost project memory

Symptoms

Wavo doesn't remember facts you established in earlier sessions.

Causes + fixes

  1. Memory opt-out — check /settings → AI preferences → Memory toggle

  2. Project memory disabledchat_project.memory_enabled = false

  3. Memory limit hitMAX_PROJECT_MEMORY_FACTS=20; oldest evicted

  4. Wrong project — opened session in different project; facts are per-project

Fix:

  • Verify memory ON: /settings → AI preferences + project settings

  • Check facts: GET /api/v1/ai-memory/facts

  • Save missing fact manually: POST /api/v1/ai-memory/facts or tell Wavo "save this as a project fact"

Issue 11: Approval card shows wrong payload

Symptoms

Wavo plans an action but the approval card shows different params than what you asked.

Causes + fixes

  1. Wavo misinterpreted intent — your prompt was ambiguous

  2. Wavo applied defaults you didn't specify — e.g. "pause adsets with bad CPA" → Wavo picked CPA threshold of 30 (default)

Fix:

  • Reject the card

  • Re-ask with explicit params: "pause adsets with CPA > 50 specifically, not 30"

Approval card is your safety net. Use it.

Issue 12: Wavo refuses to help

Symptoms

Wavo declines a legitimate question or rephrases it as a policy violation.

Causes + fixes

  1. Prompt injection defense triggered — your prompt accidentally matched a INJECTION_PATTERNS regex

  2. RBAC denial — your role lacks permission for the implied action

  3. Sensitive topic — Wavo errs on the side of caution

Fix:

  • Rephrase neutrally — avoid command words like "ignore previous instructions"

  • Verify role permission for the action

  • Provide more context — "for a legitimate troubleshooting use case..."

Issue 13: Cannot find shared session

Symptoms

Teammate shared a session but you don't see it.

Causes + fixes

  1. Wrong account / login — shared to your email; verify which email you logged in with

  2. Not in "Shared with me" view — check sidebar filters

  3. Share revoked — owner may have removed your access

Fix:

  • Check /chat → sidebar → Shared with me

  • Ask the owner to verify your email in the share

  • Re-share if revoked

Issue 14: Costs spiked

Symptoms

Credit balance dropped faster than expected.

Causes + fixes

  1. Switched to expensive model — Opus 5×/25× Sonnet

  2. Long tool chains — MAX_TOOL_ROUNDS=5 per turn; multi-turn add up

  3. Heavy file uploads — large file = many input tokens

  4. Verbose model output — output tokens often 3-5× input cost

Fix:

  • Check usage breakdown: /settings/team/billing → recent transactions

  • Switch to cheaper model for routine

  • Set tighter monthly cap

See ai-109 credits-limits cost-saving tips.

Escalation

If none of the above match:

  1. Capture: message ID + session ID + observed behavior + expected behavior + timestamp

  2. Give thumbs-down feedback with details

  3. Ask Wavo: "open a support ticket about this issue"

  4. Support team will follow up via Pylon

Related