🔎
Scan before the provider call
Detect email, phone, IBAN, VAT IDs, PESEL, national IDs, and other sensitive patterns before traffic reaches the LLM.
Open source · Apache 2.0 · Built for European teams
Evidence-grade AI governance in one URL change.
Dativo Talon sits in front of OpenAI, Anthropic, Bedrock, Azure OpenAI, and OpenAI-compatible clients. Every request is policy-checked, PII-scanned, cost-tracked, tool-governed, and written as signed evidence.
GDPR signals
NIS2-ready evidence
EU AI Act oversight
DORA-friendly audit trails
$ talon audit list
ID CALLER PII COST(€) DECISION
evt_a1b2c3 support-bot email(1) 0.003 allowed
evt_d4e5f6 hr-bot iban(2) 0.000 blocked:pii
evt_x9y0z1 eng-tools none 0.000 blocked:tool
evt_g7h8i9 ops-bot none 0.012 allowed
$ talon audit verify evt_a1b2c3
✓ Evidence signature VALID
Governed LLM traffic
PII controls
Tool governance
Cost limits
Signed evidence
Why Talon
AI adoption is easy. Proving control is hard.
Talon gives engineering and compliance teams a shared control plane: policy before provider calls, evidence after every decision.
🛡️
Enforce policy before spend happens
Block, redact, warn, rate-limit, or route requests using caller-specific policy, model allowlists, and cost limits.
🧾
Generate signed evidence
Every request produces tamper-evident evidence with classification, policy decision, model, cost, hashes, and HMAC signature.
60-second demo
See governance working without an API key.
The Docker demo uses a mock provider, but Talon still runs the real pipeline: PII scan, policy decision, cost tracking, and signed evidence.
git clone https://github.com/dativo-io/talon
cd talon/examples/docker-compose
docker compose up
curl -X POST http://localhost:8080/v1/proxy/openai/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-4o-mini",
"messages": [{
"role": "user",
"content": "My email is jan@example.com and my IBAN is DE89370400440532013000."
}]
}'
docker compose exec talon /usr/local/bin/talon audit list
docker compose exec talon /usr/local/bin/talon audit verify req_a1b2c3d4How it works
Your app keeps the same SDK. Talon becomes the governed path.
Your app
Slack bot, OpenClaw, internal tool, SaaS app
→
Talon Gateway
PII · policy · cost · tools · evidence
→
LLM provider
OpenAI · Anthropic · Bedrock · Azure · local
# Start local proxy
talon serve --proxy-quickstart --port 8080
# Point your SDK at Talon
export OPENAI_BASE_URL=http://127.0.0.1:8080/v1
export OPENAI_API_KEY=sk-...from openai import OpenAI
client = OpenAI(
base_url="http://127.0.0.1:8080/v1",
api_key="sk-..."
)
client.chat.completions.create(
model="gpt-4o-mini",
messages=[{"role": "user", "content": "Hello"}]
)Core features
More than a gateway. More than a PII proxy.
Talon combines policy enforcement, AI security, FinOps, and compliance evidence in one open-source binary.
🧾
Evidence & compliance
HMAC-signed evidence records, audit list/show/verify, CSV/JSON export, and compliance-friendly processing records.
🔐
PII & data protection
Input and response scanning, redact/block/warn modes, EU-focused identifiers, and data-tier classification.
💶
Cost governance
Per-caller budgets, daily and monthly caps, request cost attribution, model breakdowns, and budget utilization.
🛠️
Tool governance
Filter or block dangerous tools before the model can call them. Record allowed, filtered, and denied tool decisions.
🇪🇺
EU data sovereignty
Provider metadata, EU strict/preferred/global routing modes, Azure OpenAI EU regions, Bedrock EU regions, Mistral, and local Ollama.
📊
Dashboard & observability
Gateway dashboard, metrics API, SSE stream, OpenTelemetry traces and metrics, and structured logs.
Auditor-ready by default
Every decision becomes an evidence record.
Talon records what was processed, which policy applied, which model was used, what it cost, whether PII was found, and whether the record signature is valid.
| Evidence field | Example |
|---|---|
| Tenant / caller | default / support-bot |
| Policy decision | allowed:redacted |
| PII detected | email, IBAN |
| Model | gpt-4o-mini |
| Cost | €0.003 |
| Signature | HMAC valid |
talon audit export \
--format csv \
--from 2026-02-01 \
--to 2026-02-28
talon audit verify req_a1b2c3d4
✓ Evidence req_a1b2c3d4: signature VALIDIntegration paths
Start with the AI traffic you already have.
Existing apps
Change the OpenAI-compatible base URL and use a Talon caller key. Keep your SDK and request shape.
Slack and support bots
Put Talon in front of customer-facing AI workflows to add PII controls, budgets, and evidence.
OpenClaw and coding agents
Route coding-agent LLM traffic through Talon and keep an auditable trail of calls, cost, and controls.
MCP / vendor proxy
Govern tool calls and third-party AI vendor access with allowed tools, blocked actions, and redaction.
Multi-tenant MSP
Issue one caller key per customer or app, isolate evidence by tenant, and export customer-specific audit bundles.
Native Talon agents
Use talon run, governed memory, plan review, secrets, attachments, and policy-first execution.
Positioning
Where Talon fits.
Talon is complementary to AI gateways and observability tools, but its product center is governance evidence for EU teams.
| Tool type | Primary focus | EU governance | Signed evidence | Tool governance | Cost controls |
|---|---|---|---|---|---|
| Talon | Evidence-grade AI governance gateway | Yes | Yes | Yes | Yes |
| Portkey | AI gateway, routing, reliability, guardrails | Partial | No / not primary | Partial | Yes |
| LiteLLM | Provider proxy, routing, budgets | Partial | No / not primary | Partial | Yes |
| Helicone | LLM observability and traces | Partial | Partial | No / not primary | Yes |
| PII-only proxy | Request/response redaction | Partial | No | No | No |
Early preview
Put governance in front of your first AI workflow today.
Start with the Docker demo, then point one OpenAI-compatible client at Talon. No rewrite. No SaaS dependency. Evidence from the first request.