Every AI action in your company. On the record.

Route AI through Tallin and every human, app, and agent gets an owner, a budget, a policy, an off switch, and an append-only record your board, auditor, or biggest customer can inspect.

Sample Ledger

A representative append-only record. Every action, owner, and dollar, on one line.

j.alvarez · human14:02:11
claude-sonnet-5 · $0.0214OBSERVED
invoice-copilot · app14:02:13
text-embedding-4 · $0.0009ENFORCED
deploy-agent-04 · agent14:02:16
gpt-5.2 · $0.1837ENFORCED
m.okafor · human14:02:19
gemini-3-pro · $0.0442INFERRED
personal account · human14:02:22
consumer chatgpt · unknownNOT COVERED
support-triage · app14:02:25
claude-haiku-4.5 · $0.0031ENFORCED
r.tanaka · human14:02:28
claude-opus-4.8 · $0.0968OBSERVED
research-agent-11 · agent14:02:31
gpt-5.2 · $1.2406ENFORCED

Quantified outcomes

2 min
self-serve scan

See likely exposure before you talk to sales.

$2,500
paid pilot

60 days, credited against year one if you continue.

13 mo
live ledger

Then archive, with coverage labels intact.

4 states
coverage

Observed, inferred, enforced, or not covered.

Trust band

SOC 2-ready evidence
Coverage-aware exports for audit, board, and customer-security requests.
Compliance posture
Controls and evidence stay tied to the AI actor, model, cost, and policy result.
Design partner cohort
Founder-led onboarding while the product hardens with live enterprise teams.
Hosted or VPC gateway
Use hosted Tallin, or keep the request path in your own cloud boundary.

Not a ledger. A control plane.

Gateway
Route AI through Tallin to record, cap, and block in real time.
Enterprise AI Supervision
Supervise sanctioned AI tools employees already use and turn compliance records into reviewable evidence.
Actor registry
Every human, app, and agent has an owner, a budget, and an off switch.
Spend ledger
Cost per actor, model, and team, on an append-only record.
Coverage labels
Observed, inferred, enforced, or not covered, on every entry.
Evidence pack
Board and auditor-ready exports of what you can prove.

Entry 01 — The Problem

The bill is the first time most companies see their AI usage.

CF-01
A full-year 2026 AI budget. Gone in roughly four months.

Claude Code spread to ~5,000 Uber engineers far faster than forecast; the annual budget was exhausted by April. Uber now caps per-tool spend at $1,500 per engineer per month.

(The Information, 2026; Bloomberg, Jun 2026)
CF-02
What unconstrained agents cost at list price: $1,305,088.81 in 30 days.

A solo research project ran ~100 agents: 603 billion tokens, 7.6 million requests, a team of three. OpenAI covered the bill deliberately, as research. Most companies will not get theirs forgiven.

(Tom's Hardware)
CF-03
Four agents. Two locked in a recursive loop. $47,000 over eleven days.

Discovered only when a human opened the invoice. An illustrative single-engineer report; the failure pattern is structural: silent recursive spend with no kill switch.

(Engineer report via Medium)

Cost is one exposure. Data is the other.

Cost

Invisible until the bill arrives.

Uber is the proof: a sophisticated, AI-forward enterprise with dashboards and a CTO, and the spend still outran the year’s budget before anyone could react. A 200-person company has less visibility and the same usage-based pricing.

Exposure

Data leaves silently.

The founding precedent of this category: Samsung, April 2023. Three source-code leaks into ChatGPT in roughly twenty days, covering semiconductor source, test sequences, and a meeting transcript, followed by a company-wide ban (The Economist Korea, Mar 2023). Today every employee has ten AI tools, and agents act on their own.

Tallin makes both visible: observed, inferred, enforced, or honestly not covered.

Entry 02 — Coverage

Most tools imply total visibility. Tallin stamps every entry with what it can prove.

OBSERVED
Recorded directly from provider telemetry, SSO, billing exports, or sanctioned AI tools through provider compliance APIs. Provable, line by line.
INFERRED
Deduced from secondary evidence such as expense records and network logs. Reliable, and labeled as indirect.
ENFORCED
Actively governed: routed through the Tallin gateway (recorded, capped, and blockable in real time), or held by access restriction, SSO, CASB, or provider admin policy.
NOT COVERED
No reliable signal exists. Tallin says so plainly instead of pretending.

Entry 03 — How It Works

Connect sources. Route AI. Enforce policy. Export proof.

Tallin has two evidence paths: gateway control for AI you route through Tallin, and Enterprise AI Supervision for sanctioned AI tools your employees already use. Both land on the same ledger and evidence layer.

How it works schematic
Tallin system schematicGateway path: humans, apps, and agents connect through the Tallin gateway to AI models. Supervision path: sanctioned AI tools send compliance records through provider compliance APIs. Both paths write to the same append-only ledger and evidence layer.HUMANSAPPSAGENTSIDENTITIESSUPERVISED AI TOOLSSANCTIONED AI TOOLSSupervision path · compliance recordsGATEWAYidentity · policy · capsblock · recordGateway path · policy and budget enforcementMODEL API AMODEL API BOTHER AI SURFACESMODELSOUTSIDE THE GATEWAY: OBSERVED, INFERRED, OR NOT COVEREDevery request and compliance record, written downLEDGERappend-only · actor · model · cost · coverage14:02:11 j.alvarez · human chat.completion claude-sonnet-5 $0.0214 OBSERVED14:02:13 deploy-agent-04 · agent chat.completion gpt-5.2 $0.1837 ENFORCED

Entry 04 — The Ledger Spec

The record, specified.

Write model
Append-only. Entries are never edited and never deleted.
Capture
Configurable per policy: metadata only, or full content when you need it. Every entry carries actor, model, and cost.
Retention
13 months, then archive.
Tenancy
Single-tenant architecture. Your ledger shares infrastructure with no one.
Coverage
Every entry stamped: observed, inferred, enforced, or not covered.

Product facts, stated as ours. Not research claims.

Entry 05 — Anchor pricing

Start with the pilot. Size the annual plan from real usage.

Tallin is priced by covered directory users and governed AI actors. The public anchor is the paid pilot; annual plans are quoted after we know what you need to govern.

Paid pilot
$2,500 for 60 days. Credited against year one if you continue.
Growth
Custom annual plan for teams starting with hosted Tallin.
Scale
Custom annual plan for departments, apps, and agents across providers.
Complete / Enterprise
Custom plan for VPC gateway, custom retention, security review, or procurement.

No fake seat math. The pilot gives both teams a factual usage baseline before procurement starts.

Request access

Entry 06 — Index

Common questions, on the record.

06.1What does Tallin record?+
Every AI action on a covered surface: the actor (human, app, or agent), the model, and the cost, plus content at the fidelity you set, from metadata only to full prompt and response. Entries are append-only; nothing is edited or deleted after the fact.
06.2What can Tallin actually see?+
Exactly what it stamps. Observed means direct telemetry. Inferred means secondary evidence. Enforced means actively governed, through the gateway or an access-restriction policy. Anything else is marked not covered, in plain sight.
06.3Does Tallin sit in the request path?+
Only where you route through the gateway; that traffic is enforced. Other coverage comes from provider telemetry, SSO, billing, and network signals, and is stamped observed or inferred accordingly.
06.4Can Tallin cap or block spend?+
Yes, at the gateway: per-actor caps, budgets, and blocks apply to traffic Tallin carries. Tallin does not claim to block traffic it does not carry; that traffic is recorded and labeled instead.
06.5How long is the ledger retained?+
Thirteen months in the live ledger, then archive.
06.6Is my data shared across customers?+
No. Tallin is single-tenant: your ledger, your infrastructure, no co-mingling.
06.7How is Tallin priced?+
By covered directory users and governed AI actors. Start with a $2,500 paid pilot for 60 days, credited against year one if you continue, then move to a custom annual plan sized to your AI footprint.

Join the design partner cohort.

Tell us about your AI actors, provider stack, and the runaway spend or policy gaps you need to close. We will help you decide whether Tallin is a fit for the first cohort.

Tallin | The Control Plane for AI Use by Humans, Apps & Agents