Aperture
coming soon
Open source · AGPL · runs in your AWS

Know what your coding AI actually costs. Per PR. Per ticket. Per shipped feature.

An open-source gateway that routes every coding agent through one governed endpoint on Amazon Bedrock, in your own AWS account. SSO, quotas, policy routing, and cost attribution that reaches the work.

Coming soon on GitHub.
The repository opens with the first public release under AGPL.
Enterprise, implemented by DNX

Built and sponsored by DNX Solutions, an AWS Premier Consulting Partner.

aperture · cost dashboard
last 30 days
Spend · mtd
$48,210
Requests
1.24M
Cost / merged pr
$37.10
Reconciled
99.4%
works with Claude Code Codex CLI OpenCode more coming soon
01 / the problem

Coding AI spend is exploding and nobody can see it.

Engineers use personal keys across a dozen tools. Finance sees one number with no breakdown. Nobody can answer "what are we getting for this?" No SSO. No quotas. No policy control.

llm proxies

LLM proxies count tokens.

They meter traffic but never tie a dollar to a person, a repo, or a result.

analytics

Analytics tools watch outcomes.

They see what ships but sit outside the request path. No quotas, no routing, no cutoffs.

aperture

Aperture does both.

Control and attribution in one gateway. Open source, in your account. No data platform required.

02 / capabilities

Enterprise-grade governance, without the platform.

01

Single governed endpoint

Every agent routes through one gateway in your AWS account. No data leaves your infrastructure.

02

SSO, no shared keys

Engineers sign in with existing identity. No personal or shared API keys anywhere.

03

Per-user & team quotas

Soft and hard spend limits, rate limits, and automatic cutoffs.

04

Policy-based routing

Route by task, cost ceiling, or org policy. Allowlist models per team, with fallbacks.

05

GitHub & Jira integrations

Link sessions to PRs and issues. Measure against merge rates and cycle time.

06

Service accounts & chargeback

Machine callers get their own keys, quotas, and tenant tags, so per-customer AI cost is a line item.

07

Dashboards & audit trail

Ships with dashboards and logs who used which model, when, and for what.

08

Nothing to run but AWS

One CDK stack. No data platform, no vendor tenancy, no egress out of your account.

09

Open source, AGPL

Read the routing logic that spends your budget. Fork it if your policy needs something else.

03 / attribution

Attribution that reaches the work.

Every request is stamped with user, repo, branch, and ticket. Spend breaks down by pull request and ticket, so you see cost per shipped feature, not cost per token.

user repo branch ticket PR
coverage · 30d
attributed $1,046.00 unattributed $168.83
Unattributed spend is always shown beside attributed spend. The only open-source tool connecting AI spend to PRs and tickets.
Cost breakdown
Grouped by the dimension you're accountable for.
PAY-1183Streaming tokenizer
$312.40
4,210 reqClaude Sonnet 4.5Merged
INF-902Retry backoff
$148.10
1,980 reqClaude Opus 4.1Merged
SEC-77Auth guard refactor
$96.70
2,640 reqClaude Sonnet 4.5In review
SRCH-12Vector cache spike
$58.30
5,120 reqGPT-5.6 TerraOpen
OBS-40Trace sampling
$21.15
640 reqClaude Haiku 4Merged
04 / service accounts

Not every request comes from an engineer.

If your product is multi-tenant and uses AI, give each application, batch job, and tenant its own service account: its own key, its own quota, its own tags. Aperture attributes spend to the customer that caused it, so you can charge it back, price a plan, and find the tenant that costs more than it pays.

Spend by tenant
One service account per tenant. Same gateway, same quotas.
august 2026
tenant requests spend plan
Acme Corp
svc/doc-summarizer
18,420 $412.80 Scale
Northwind
svc/inbox-triage
11,260 $268.10 Scale
Globex
svc/contract-review
6,940 $151.35 Team
Initech
svc/inbox-triage
4,180 $96.20 Team
Trials (pooled)
svc/sandbox
2,050 $18.20 Free
Every row exports as a line item, ready for your billing system.
Tag a request, get a line item.

Two headers on your own outbound call. No SDK, no sidecar.

Request
curl https://aperture.internal/v1/messages \ -H "authorization: Bearer $APERTURE_SERVICE_KEY" \ -H "x-aperture-tenant: acme-corp" \ -H "x-aperture-workload: doc-summarizer" \ -d @request.json
Quotas apply to machines too.
svc/doc-summarizer $4,120 of $6,000
svc/nightly-reindex $1,880 of $2,000
A service account that hits its hard limit stops. It never silently spends a customer's margin.
05 / architecture

Runs in your AWS account. Zero egress.

Traffic from every agent flows through one gateway you own, then to Amazon Bedrock. Aperture never sees your data, because it runs inside your boundary.

coding agents
Claude Code
Codex CLI
OpenCode
service accounts
Your multi-tenant app
Batch & scheduled jobs
sso · keys
your aws account
Aperture Gateway
quotas routing attribution audit
models
Amazon Bedrock
Selected models — Claude, GPT, Llama, Mistral, or whatever your policy allowlists
06 / editions

Start open source. Scale with DNX.

open source · agpl the hero

Aperture Community

Complete for a single team on a single AWS account. No feature gates, no sign-up walls.

Governed endpoint on your Bedrock
SSO, quotas, policy routing
Attribution to PRs and tickets
Service accounts and per-tenant chargeback
Dashboards and a full audit trail
coming soon on github
enterprise

Implemented by DNX

Everything in Community, plus multi-account governance implemented per customer by DNX.

Multi-account / multi-org
SAML (Okta / Entra)
Audit log export
Advanced analytics & policy packs
Talk to DNX