toklume Contact us

In development · Independent project

Give every agent
a token budget.

A clearer view of where tokens go, and a considered limit for the next request. Toklume is being built for developers working with AI agents.

Explore the concept

Planned first release: usage by run, budget warnings, and request checks.

THE IDEA, AT A GLANCE01 / 03
AGENT REQUESTresearch-agent / next step
One place to count.
One place to decide.
Account for usage
Check the budget
Allow the next request
MODEL PROVIDERConcept illustration
Visibility by agentBudgets in tokensA decision before the next request

01 — THE BUDGET EXPLORER

Make the boundary visible.

Pick a sample workflow and adjust its budget. See how recorded usage and estimated next steps fit together.

Interactive conceptSAMPLE DATA · RUNS ENTIRELY IN YOUR BROWSER
AGENT LEDGER

research-agent

SAMPLE / 001
Recorded usage12,400 tokens
Projected total16,300 tokens
Room in budget3,700 tokens
81.5% projected20,000 token budget
  • Recorded
  • In flight · estimate
  • Next request · estimate
Warning threshold reached

The sample projection crosses 80% of the budget. A warning would help you review the next step.

In-flight estimate 2,100 tokensNext-request estimate 1,800 tokens

A sample of the planned controls, running locally with illustrative data. Projected usage includes estimates and may differ from actual usage.

02 — A SMALL PIECE OF YOUR STACK

Between an agent
and its next request.

The planned first version focuses on a single provider and a straightforward accounting loop. Less to integrate. Less to interpret.

01 / IDENTIFY

Give the run a name.

Tag requests by agent and project so usage has a clear owner and a useful context.

agent: "research-agent"
run:   "brief-042"
02 / ACCOUNT

Keep a token ledger.

Record provider-reported usage, with estimates kept visibly separate while a request is in flight.

recorded: 12_400
in_flight_estimate: 2_100
03 / DECIDE

Check the next step.

Warn as the budget approaches its threshold. Plan to block new requests once the recorded limit is reached.

budget: 20_000
warn_at: 0.80

A budget is only as timely as its accounting.

Streaming and concurrent requests can consume tokens before final usage is available. The planned controls apply to subsequent requests; they do not guarantee an exact token ceiling or fixed dollar spend.

03 — STARTING SMALL

A focused first
version.

Toklume is a pre-launch project. The roadmap below shows the focus for the first version.

  1. 01

    One provider. A useful ledger.

    Request tags, provider usage records, and a simple view of tokens by run.

    PLANNED
  2. 02

    Thresholds that are easy to reason about.

    Token-budget warnings and checks before subsequent requests.

    PLANNED
  3. 03

    A small, inspectable integration.

    A gateway or SDK entry point, basic exports, and documented accounting limits.

    PLANNED

BUILDING WITH DEVELOPERS IN MIND

What would make
your token usage clearer?

Questions, a difficult workflow, or a perspective on the idea? Get in touch.

Contact us hello@toklume.app