Treeline Proxy
Runtime Trust Broker for GenAI Traffic
deterministic enforcement • proof artifacts • privacy-safe telemetry

Secure GenAI traffic without slowing teams down.
Control at runtime. Prove it after.

Treeline Proxy enforces signed, versioned policies on prompts, responses, and agent tool calls — producing deterministic allow / redact / block decisions and audit-grade, privacy-safe telemetry your security team can defend.

Signed policies
ed25519 + versioned
Deterministic
allow/redact/block
Proof-grade
test vectors + CI evidence
Gateway-native
fits your stack
Built for real governance reviews
“We needed something we could explain to a CISO and prove to an auditor — without retaining prompts.” — platform security lead (pilot conversation)

GenAI has escaped the control plane

Security teams can’t govern prompts like code. Platform teams can’t ship AI if they can’t prove control. “Trust us” does not pass audits.

🧨

Prompt leakage bypasses DLP

Prompts and tool calls often never hit traditional controls. Secrets and PII slip through invisible paths.

  • Credentials in prompts
  • PII in outputs
  • Regulated terms
🧰

Agents create unlogged actions

Tool calling expands blast radius: actions happen outside the normal app boundary.

  • Unapproved endpoints
  • Data exfil routes
  • Shadow workflows
🧾

No proof of enforcement

If you can’t show what policy ran and what it did, you don’t have governance — you have hope.

  • No policy version trace
  • No evidence artifacts
  • No audit narrative
Hard truth
If you can’t prove control over LLM traffic — prompts, responses, and tool calls — you don’t have control.

Want a low-friction starting point?

Start with a pilot that produces a signed policy pack + evidence artifact your team can review.

Three deliverables: Policy, Proof, Telemetry

Deterministic enforcement you can defend to executives, auditors, and counsel.

🔐

Policy

Signed policy packs that evaluate GenAI payloads consistently — across teams and vendors.

  • ed25519 signatures
  • Monotonic versions (vX.Y.Z)
  • Rollback-safe releases
🧪

Proof

Test vectors + CI evidence artifacts: a verifiable record that policy vX did what it claims.

  • Golden test corpus
  • Expected results
  • Compliance-ready outputs
📊

Telemetry

Privacy-safe metrics and dashboards that show control without retaining raw prompt content.

  • Decision rates
  • Top rule hits
  • Latency SLOs

Get architecture + pricing

Deployment patterns (Envoy / API GW / CloudFront + WAF) and commercial tiers designed for pilots → regulated production.

Proof you can show a CISO

Every decision is traceable to a signed policy version. Evidence is machine-verifiable.

Signed policy pack (example)

policy-pack-v1.2.0/
├─ rules.yml
├─ thresholds.yml
├─ tests/
├─ manifest.json (sha256, issuer, version)
└─ signature.sig (ed25519)

Policy packs ship like code. Rollouts are measurable. Rollbacks are safe.

CI evidence (example)

✅ verify_signature: PASS
✅ tests: 312 passed, 0 failed
✅ expected_results: MATCH
✅ publish: v1.2.0 (immutable)

artifact: evidence-v1.2.0.zip

This is what turns “we think it’s safe” into “we can prove it.”

Why Treeline exists
Security without proof is theater. Treeline is built to produce the artifacts that win governance reviews.

Privacy-safe by default

Governance without raw prompt retention. Keep sensitive text out of logs while keeping decisions provable.

🕳️

No raw retention

Default posture is no raw prompt storage. Export derived signals only (decision, rule hits, counters, latency).

🧾

Audit narrative

Decisions are traceable to a signed policy version, with test vectors and evidence artifacts for reviewers.

🧩

Compliance-ready posture

Designed to support regulated environments. Add-on reporting and approvals available in regulated tiers.

Have a compliance deadline?

Tell us your framework and we’ll map policy packs + evidence outputs to what auditors ask for.