For teams building with AI
One tool for your AI's resilience, cost, quality and security.
ZokLens watches every request your AI makes.
Connect it with one line of code.
This is an early preview of what we are building. Sign-up opens at launch. Until then, write to us any time. [email protected]
The product
One trace. Eight views. Ask about any of them.
Every call and agent run lands in ZokLens once. Each module reads the same trace. Copilot answers questions across all of them, and proposes changes that apply only after you confirm.
Why did our AI spend go up last week?
Spend rose 38% ($1,092 → $1,507). $412 of the $415 increase came from Support: the ticket-summary agent switched from gpt-4.1-mini to gpt-4.1 on Oct 2.
Last 7 days by team
October forecast
$6,180 – $6,940 at current trend
Setup
Start with one line. Go deeper when you need to.
Each step adds detail. You can stop at any of them.
1Point your client at ZokLens
Change base_url and put your ZokLens key in front of your provider key. Every call is traced and priced.
2Add the SDK
Group calls into agent runs and sessions. Spans are sent in the background.
3Or let your coding agent do it
Copy a ready-made setup prompt from Integrations and paste it into the coding agent that works on your app.
from openai import OpenAI client = OpenAI( api_key="zok_live_…:sk-…", # ZokLens key : your provider key base_url="https://api.zoklens.com/v1", )
# pip install zoklens import zoklens zoklens.init( api_key="zok_live_…", endpoint="https://api.zoklens.com", project="refund-agent", ) zoklens.instrument(providers=["openai", "anthropic"])
OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=\ https://api.zoklens.com/api/v1/otel/v1/traces OTEL_EXPORTER_OTLP_HEADERS=X-API-Key=zok_live_… OTEL_SERVICE_NAME=refund-agent
# Integrations → API keys → Copy prompt # "Paste this into the coding agent that will instrument your app." - ZokLens API origin — the official SDK's endpoint: https://… - Auth header: X-API-Key: zok_live_… - service.name: replace-with-your-service-name - Do not silently auto-upgrade SDK dependencies in production. …
Sound familiar?
Six questions every team running AI gets asked.
Each one starts with a moment most teams have lived through. Each answer comes from the same traces.
At 9 a.m. your main provider starts returning errors. Your support bot goes down with it.
What happens when a provider fails?
- Your app calls a named route, such as support.chat. ZokLens holds the model behind it.
- Switch the model or provider in ZokLens. No redeploy.
- Turn on failover: on a 429 or 5xx, ZokLens tries your standby, up to three steps
- See provider slowdowns and error spikes from your own traffic
Finance asks: “We spent $47,000 on AI last month. What did we get for it?” You open your provider dashboard and see total tokens. That's it.
What did AI cost us, and who spent it?
- Spend by model, department, team, project and API key
- Month-end forecast with a confidence band
- Projected month-end spend against your budget, with alerts
- Chargeback reports for each department
| Department | Top model | Spend | vs Sep |
|---|---|---|---|
| Support | gpt-4.1 | $2,840 | +41% |
| Search | claude-sonnet | $1,512 | −6% |
| Sales | deepseek-chat | $1,061 | +3% |
| Internal tools | gemini-flash | $688 | −12% |
A customer says the assistant “just hangs”. One request took 9 seconds, and your logs only show that it finished.
Where did the time go?
- A span tree for every trace: each model call, tool call and step
- Tokens, latency, errors and cost on every span
Your agent ran for twenty minutes overnight. It finished, but it cost ten times its usual run.
Why did that agent run go wrong?
- Repeated-call loops and stuck runs
- Error spikes and cost anomalies for each agent
- Provider slowdowns, seen from your own traffic
- A map of which agents call which models and tools
refund-agent · web_search called 6× with the same input
research-agent · no progress for 9 min
ticket-summary · 3.4× its 7-day median
anthropic · error rate 4.1% over 15 min
Nothing errored. The calls returned 200, the tokens were billed, the pipeline “worked”. The answers were empty.
Are our answers getting worse?
- Flags answers that came back 200 but empty, truncated or refused
- Tells you when a source goes quiet
- An AI judge scores relevance and groundedness, on your own provider key
- Flagged traces land in a review queue for a person to confirm
A support agent pastes a customer's social security number into your chatbot to look up their account.
Is customer data reaching the model provider?
- Zokforce Guard masks personal data before the request is forwarded
- Emails, phone numbers, card numbers, SSN/SIN, IBANs, IP addresses and medical identifiers
- Real values are restored in the response, so users see no difference
- Dry-run mode shows what would be masked before you turn it on
Compliance
ProShows the evidence your AI traffic leaves, for any department or team, and names each gap. Evidence, not a pass/fail verdict. Export a pack as PDF or ZIP for your auditor.
Product Analytics
ProWhen you switch models on the same job, compare them with a significance test before you call a winner.
Zokguard
Add-onGuard Premium: a dedicated managed Guard runtime that adds secret and prompt-injection blocking. Add it to any paid plan.
Who it's for
One workspace. Each team sees what it needs.
Your org tree decides who sees what: people see their own unit and below, never a sibling team.
Debug the call, not the logs
- Span tree for every trace
- Python and JS/TS SDKs on OpenTelemetry
- Playground: compare models side by side on your own key
Keep every integration in bounds
- Named routes with a standby provider
- Rate limit, monthly token cap and monthly budget on each key
- Provider health from your own traffic
- Alerts in-app or by email, chosen per alert type
Answer finance in one screen
- Spend by department, team and project
- Month-end forecast with a confidence band
- Chargeback reports
Show evidence, not promises
- Personal data masked before it is forwarded
- Append-only audit log, exportable to CSV
- Fields visible, masked or hidden per role
Know if answers got better
- AI judge scores for relevance and groundedness
- Model comparisons with a significance test
- Review queue for flagged answers
Security
What happens to your data.
ZokLens stores prompts and responses so you can inspect them. Here is how that is controlled.
Mask before it leaves
On Pro, Zokforce Guard masks personal data before a request is forwarded to your provider. Dry-run first to see what it would catch.
Keys handled carefully
Provider keys you store are encrypted. ZokLens API keys and access tokens are stored only as hashes.
Access you control
Workspaces are kept separate. Roles, custom roles and an org tree decide who sees what, down to individual fields.
A record of every change
An append-only audit log of configuration changes and actions, exportable to CSV.
Retention you set
30, 90 or 365 days by plan, configurable per resource. Expired data is cleared by a daily job.
Leave whenever you want
Export everything as a ZIP of JSON and CSV. Closing your account starts a 30-day grace period you can reverse, then a full purge.
Where we are on audits: Designed with SOC 2 readiness access controls. Privacy-rights workflows are implemented for readiness. We are not certified yet, and this page will say so when we are.
Questions
What teams ask before they sign up.
Proxy or SDK: which should I use?
The proxy is the fastest start: change one line and every call is traced and priced, and Guard can mask data before it is forwarded. The SDK adds agent runs and sessions and sends spans in the background, outside your request path.
What happens if ZokLens goes down? Does my app break?
With the proxy, your LLM calls go through ZokLens, so they fail while it is unreachable. If ZokLens can't save a trace, the call returns an error instead of going unrecorded. With the SDK or OpenTelemetry, spans are exported in the background. If export fails, your LLM calls carry on.
Do you store my prompts?
Yes. Prompts and responses are what you inspect in traces. They are kept for your retention period and can be exported or deleted. On Pro, Guard masks personal data before it reaches your provider.
Can I keep my own OpenAI or Anthropic account?
Yes. Put your ZokLens key in front of your provider key (zok_live_…:sk-…), or store the provider key once in ZokLens, encrypted. Your provider bills you at your own rates.
Can I export my data and leave?
Yes. Export everything as a ZIP of JSON and CSV, delivered through a signed link that works for 48 hours. Closing your account starts a 30-day grace period you can reverse, then all data is purged.
Is ZokLens SOC 2 certified? Do you support SSO?
Not yet, for either. ZokLens is designed with SOC 2 readiness access controls, and access today is managed with roles, custom roles and field-level permissions.
Which providers work?
OpenAI, Anthropic, Google Gemini, DeepSeek, Mistral, Groq, Together, xAI Grok, Cohere, Azure OpenAI, AWS Bedrock, Qwen, Baidu, Zhipu, Moonshot, MiniMax, SiliconFlow and Cloudsway, plus any OpenAI-compatible endpoint through the SDK.
What ZokLens is not
- It doesn't change your agents. It records, flags and recommends. Models switch only on routes you set up.
- It isn't a general APM. It watches your LLM calls and agent runs, not your whole infrastructure.
- It doesn't host or fine-tune models. Your calls go to the providers you choose, on your keys.
About us
Built by Zokforce.
Zokforce Technologie Inc. is an AI company incorporated in Canada in 2025. We help businesses put AI to work: strategy, LLM integration, chatbots and process automation.
Every AI project we delivered raised the same questions: what does it cost, are the answers good, and is the data safe? ZokLens is our answer to all three, in one tool.
- Customer success
- Ethical & responsible AI
- Relentless innovation
- Honesty & transparency
“To democratize AI technology and make digital excellence accessible to businesses of all sizes.”Zokforce mission
- Incorporated
- 2025, Canada
- What we do
- AI consulting, LLM integration and automation
- Products
- ZokLens, Medscribe AI
- Contact
- [email protected]
These government qualifications are held by Zokforce Technologie Inc., the company behind ZokLens.
AI Source List, Bands 1, 2 and 3
Qualified through Public Services and Procurement Canada's Invitation to Qualify (solicitation EN578‑180001/B), in the top qualification tier.
Vendor of Record for AI Scribe
We worked with Supply Ontario to qualify our Medscribe AI as a Vendor of Record under Tender‑20123 (Artificial Intelligence Solutions – AI Scribe). It is one of only 30 vendors pre-qualified, covering the Ontario Public Service and public healthcare sector from 2025 to 2028.
ZokLens launches at the end of 2026. See you then.
This is an early preview of what we are building. Sign-up opens at launch. Until then, write to us any time.
Rolling ZokLens out across teams?
Write to the people who build it and we will set up 20 minutes. No sales script.
Email us to book