One install, four environments
The same Howlet runs on a developer's laptop, in an offline factory, and on AWS. You don't relearn the product when the environment changes.
From your model to your data, one platform.
One install, four environments, twelve modules. Your data stays with you.
Not a marketing claim. The product's verified scope.
12Modules
4Deployment modes
30 daysStatutory erasure SLA
7Systems erased per request
Why Howlet
Not slogans. Architecture decisions.
Every headline below is a real mechanism your engineers can inspect on the product page.
Howlet Coding Agent
A natural-language command connects to your own code repo and builds AI and data flows in Howlet Studio 50-100× faster. Your source code never leaves; critical actions require human approval, and every step lands in a tamper-evident audit trail.
GDPR erasure is not a single SQL DELETE
When a subject requests erasure, the data is removed in cascade across seven separate systems. The 30-day window is monitored live, and a signed evidence pack is produced for the regulator.
Your data stays in Turkey
Deploy to AWS Istanbul Local Zone and keep all data within Turkish borders. KVKK data residency is satisfied at the infrastructure level, not through contractual workarounds.
Coding Agent
Builds autonomously inside your own repo
Send a natural-language command and the Howlet Coding Agent connects to your own code repository, building AI and data flows for you. Your sensitive source code never leaves.
How it works
Connects to your own repo
GitHub, GitLab, Bitbucket or Gitea — the agent runs against your Git infrastructure. Code never leaves the perimeter.
Natural-language chained actions
One prompt: 'Prioritize last week's critical IT tickets and escalate unresolved ones to managers.' The agent chains the steps, executes them, reports back.
Role-based access control (RBAC)
Which systems the agent can touch, which resources it can read, all defined ahead of time. It cannot step outside its scope.
Human-in-the-loop for critical actions
Production deployments, data deletion, high-impact operations: approval is mandatory. Agent speed combined with human control.
Tamper-evident audit
Every command, every tool call, every file change recorded in a SHA-256 hash chain. Retroactive tampering is impossible.
50-100×faster, lower cost and higher quality than classic development cycles
Pick your own model
Stay fully on-prem for sensitive projects; tap external providers when performance is the priority.
Local LLM
Llama, Mistral, Qwen and similar open-weight models on your own GPUs. No data leaves the organization.
Cloud providers
Claude Opus 4.7, GPT-5.5 Codex, Bedrock profiles and more. Swap models per workload.
Integrated with every Howlet module
The Coding Agent is not a standalone tool. From data pipelines and model training to MLOps and user authorization, it operates under the same RBAC, the same audit trail and the same identity across every Howlet module.
Data pipelinesModel trainingStream processingKnowledge baseRelease managementSource control
12 modules
One codebase, one permission model, one audit trail
Replace 6–8 vendors in your AI/ML stack with a single platform; clear your integration backlog.
Agents · MCP · Guardrails
GenAI Studio
Build production-grade AI agents. Each agent runs isolated, with three layers of safety filtering on input, output and execution.
Natural language · Repo-connected
Coding Agent
Send a natural-language command and the Howlet Coding Agent builds AI and data flows 50-100x faster. Your source code never leaves; every step lands in a tamper-evident audit trail.
RAG · Knowledge base
DataHub
Open your enterprise documents to AI safely. 11 connectors, four vector backends and Turkish-aware PII detection out of the box.
Visual pipelines · auto-engine
Data Engineering
Design ETL and transformation flows visually. Howlet picks the right engine: fast on small data, Spark at scale. You never choose the engine manually.
Training · HPO · Eval
ML Workshop
Train models, run hyperparameter sweeps, measure fairness. GPU quotas, idle shutdown and audit logging are built into every step.
Registry · Deploy · Drift
Model Ops
Register models, promote across three environments, choose canary, shadow or A/B. When drift is detected, retraining is triggered automatically.
Real-time · exactly-once
Stream Processing
Kafka, Flink, Spark or NiFi. Howlet picks the engine that fits your topology, not the other way around. All designed in one canvas.
Git · GitOps
Source Control
Your visual artifacts sync with Git automatically. A single atomic commit pushes 30 files to any of four major Git providers.
Metrics · Trace · Correlation
Observability
Trace a user request from sign-in to LLM call in one window. A log line, a distributed trace and an AI invocation share the same correlation ID.
GDPR · KVKK · Policy
Governance & Compliance
Day-one operations for erasure requests, consent management and policy enforcement. The file you'd hand a regulator is one click away.
Identity · MFA · SSO
IAM
Let employees sign in with their existing corporate identity; when permissions change, they don't need to log in again. SSO and MFA are standard.
Authority · License · GPU · DR
Administration
Four authority tiers: root, platform, tenant, workspace. Idle GPUs scale to zero; backups run continuously without disrupting users.
Erasure request DSR-2026-0042
30-day SLA · Day 4LIVE
Systems erased
5 / 7
Hash chain
OK
- 01Platform PostgresDone
- 02Tenant PostgresDone
- 03OpenSearchDone
- 04ClickHouseDone
- 05LangfuseDone
- 06ValkeyRunning
- 07MinIOPending
Signed evidence pack, ready for the regulator.
Governance & Compliance
Erasure is not a single click. It is a process you can show to a regulator.
When a subject requests erasure, Howlet cleans every system the data passed through. The 30-day statutory window is tracked live and a signed evidence pack is produced automatically.
- Seven-system erasure cascade: platform DB, tenant DB, search indexes, analytics, LLM trace, sessions, object storage.
- Consent receipts signed with ISO 27560 JWS; retroactive policy mutation becomes impossible.
- Audit rows linked by SHA-256 hash chain; every row carries the previous and current hash.
- Regional PII enabled by default: Turkish national ID, tax ID, IBAN and phone, all with checksum validation.
- SIEM fan-out built in: Splunk, QRadar and Sentinel via CEF, HMAC-signed, queued retry on failure.
- AWS Turkey Local Zone support: deploy to Istanbul so data never leaves Turkish borders. KVKK data residency satisfied at the infrastructure level.
Deployment
Install it where you need it. Same package, same logic.
Four deployment modes, four capacity profiles. Architecture stays the same; only capacity changes. Air-gapped customers get a single USB bundle: images, configuration and offline license included.
4 deployment modes
On-Prem K8S
Single-node POC
Kubernetes
Air-Gapped
USB bundle
Offline install
AWS Single-Node
EC2 + k3s
TR Local Zone ready
AWS EKS
EKS + ECS
Multi-AZ HA · TR Local Zone
4 capacity profiles
Smallk3s single-node
16 GB · 1 user · 1 node
POC, dev, small team
Mediumk3s/EKS HA
192 GB · 100 users · 3-node HA
Departmental platform
LargeEKS / on-prem
320 GB · 500 users · 5+ nodes
Enterprise core
EnterpriseMulti-AZ DR
896 GB · unlimited · 7+ multi-AZ
Bank, telco, public sector
Profile upgrades involve no data migration: a single command adds nodes, brings up storage and updates configuration. No downtime for users.
Who gets what
A concrete answer for each role
Engineering leaders, CTO/CIOs and CISO/DPOs. Three audiences, three answers in their own language.
Engineering Leader
VP Engineering · Head of ML · Head of Data Platform
- Collapses 6–8 vendors in your stack into one and clears your team's integration backlog.
- Closes the enterprise gaps in open-source tooling (multi-tenancy, RBAC, audit) on your behalf.
- Batch, streaming, ETL and agents in one visual canvas. The platform picks the engine.
CTO / CIO
Strategic technology leadership
- The same product runs from laptop to multi-AZ; no vendor swap or retraining when scope changes.
- Live model endpoints continue to serve for 30 days even if the license isn't renewed; customers don't notice.
- Capacity upgrades happen without data migration. You scale up with no downtime.
CISO / DPO
Compliance · Risk · Privacy
- Respond to an erasure request within 30 days, across seven systems, with a signed evidence pack.
- Audit trail is hash-chained; auditors can verify integrity without touching the database.
- Turkish PII (national ID, tax ID, IBAN, phone) on by default. Market priority is in the code, not in translation.
What's new
Active development, transparent release flow
Every release is planned against 60+ ADRs under mds/stories and docs/design. Every change lands in audit.
Read the release notesv1.0.0
March 2026
- GenAI Studio Code Agent: persistent sandbox runtime with warm pool, spot failover and session resume.
- Stream Processing v2.0: five-engine portfolio, blue/green deployment and built-in schema registry.
- DataHub: Turkish PII recognizers default; four vector backends with hot swap.
- Model Ops: four-test drift ensemble; automatic retraining when thresholds cross.
- Profile upgrades with a single command: `howlet-admin profile upgrade --to medium`.
Sovereignty & compliance
Your data stays on-prem, in your air-gapped facility or inside your own AWS account. Howlet never ships telemetry off your perimeter.
GDPR Art. 15/17/20
KVKK Art. 7
ISO 27560 Consent
SHA-256 hash audit
TCKN / VKN default
SAML / OIDC / LDAP
AWS TR Local Zone
Air-gap parity
SIEM CEF fan-out
Ready when you are
From model to product in 30 minutes, with a demo tailored to your scenario.
Training to serving, RAG to streaming, RBAC to disaster recovery. Not vendor lock-in. Installed on your own infrastructure.
On-prem · air-gapped · AWS. Your data stays with you, the proof stays in the chain.