Skip to content
Trazum

Trazum

Roadmap

Where the project is going, in order, with the work that belongs to each step. What is not scheduled is written down with the reasoning.

There are no dates on this page, on purpose. The order is the commitment and the calendar is not.

In order

The first is next. Each title opens the live milestone on GitHub, which holds the same issues.

An issue shows its title when the repository’s table of open work has it, and its number otherwise.

  1. 01

    2.7.13: repeated-task evaluation

    Validate the instruction policy over repeated experiments on accepted tasks before it becomes a default: cost per accepted task, quality, retries and controller overhead, measured together.

  2. 02

    A site that carries the brand

    The landing as the front door, the documentation, the catalogue of models and prices, the roadmap and the changelog as pages of one site, and a brand page, all in the design system the project already chose and each with a way to GitHub.

  3. 03

    Trazum speaks first

    The status line, the watcher and the spend guard exist and are off until somebody wires each by hand. Next: one command that turns them on and off, a watcher that reads the agents' sessions directly and says so on the screen, a weekly summary that arrives on its own, a plan that can open the pull request, and a terminal that carries the brand.

  4. 04

    3.0.0: the release allowed to break

    A major is the only release that may remove something. This one removes `tier` from a model in the pricing catalogue, deprecated since 1.8.0 in favour of `capability`, and nothing its changelog entry does not list.

    2 issues

  5. 05

    The price anyone can check

    Every comparable tool prices from the same community file and calls the figure an estimate, and a model it cannot price counts as zero. Next: the catalogue published as an open dataset with the provider's page and the date read behind every row, and a pull request comment that says what a change does to a prompt's monthly cost against the base branch, each figure with its dated price.

    2 issues

  6. 06

    The bill where a team already looks

    The bill is printed, saved or gated. Next: the same measured figures as metrics for the dashboards a team already runs, and what was spent in a repository while a branch was written, said as a time window and never as the cost of the change.

  7. 07

    A repository a stranger can contribute to

    What a visitor and a contributor need that is not the product: a roadmap that opens with what comes next, a wiki that publishes itself, documents that agree with each other, and Discussions with a form for each kind of question.

  8. 08

    Tests that would notice a wrong price

    Line coverage says the code that prices a call runs under test. Mutation testing says whether a test would notice it pricing a call wrongly, and today it reaches six files of the core and, in two more, the functions that choose a rate and multiply by it. Next: the catalogue's own numbers, where a sample found published prices that no test would notice changing.

  9. 09
  10. 10
  11. 11
  12. 12

    Where people meet Trazum

    The editor extension on the VS Code Marketplace, the playground as a Hugging Face Space, the hosted web app measured for how fast it loads, off unless its operator asks, a page that says what Trazum Pro is and that it is not open yet, and a download page at /download for the files each release carries.

  13. 13

    The optional model pass

    Choose and test a model-side optimiser candidate, behind the opt-in layer, with the deterministic core unchanged.

Under consideration

Not scheduled, and kept so the reasoning is on the record.

Not scheduled. Listed so the reasoning is on the record.

Nothing here is built, apart from the entries struck through or annotated as shipped, which stay as the record of what this file went back on. That is the whole meaning of the heading, and it stopped being true without anybody noticing: the editor extension shipped as trazum-vscode in 1.86.0 and the per-model-family tokenizer as @trazum/tokenizer-openai in 1.85.0, while both stayed listed below as things this file was still thinking about. The release sections in HISTORY.md recorded them as released the whole time, so one file said both at once. roadmap-truth.test.js holds it now, because the moment something is built is exactly the moment nobody remembers to move it.

  • More contradiction axes — tone, persona, refusal policy. Unscheduled rather than planned, because each new axis has to earn its place against false positives and I have not seen enough real prompts to know which ones do. An advisory people learn to ignore is worse than no advisory, and four axes that fire correctly beat seven that mostly do not.

  • More locales. The architecture supports it as of 0.3.0, and adding one is now a catalogue plus dictionary entries. Held back on purpose: a language needs a maintainer who actually reads it, and a stale translation is worse than an honest fallback to English.

    What that role is was written down at 1.56.2, with its bar, in docs/language-maintainer.md — along with the admission that five of the seven dictionaries already shipping had never been read by a speaker either. The page does not move this entry into the schedule. Whether somebody fills the role is not something this project can plan.

    Japanese is deliberately not on this list, and the split is the point. There is no Japanese trimming dictionary and one is not planned: deciding that a phrase says something in more words than it needs is a judgement about the language, and nobody here can make it in Japanese. But --reorder's backward-reference list does cover Japanese and Chinese, matched without word boundaries because 上記のテキスト has none. Those two are not the same claim. Refusing to rearrange somebody's prompt needs only enough of the language to recognise a phrase pointing backwards; offering to shorten it means asserting the shorter version still asks for the same thing. Trazum will do the first in a language it cannot read and will not do the second.

    So a Japanese prompt gets --reorder's protections in full, no trimming, and a report that names the seven languages the dictionaries cover rather than implying the prompt was already efficient.

    Still unscheduled, and the number that was missing now exists for two families out of six. Pricing seven providers made this the question it always was in miniature: the estimator is tuned against Claude's tokenizer, and a GPT or Kimi figure carries whatever error that mismatch produces. This entry set the threshold itself — within 5% across families and the dependency is not worth taking; 40% out and it is — and said deciding without that number would be deciding blind.

    1.82.0 measured it. Against DeepSeek's own counter the estimator is 94.5% out at worst over the 47-sample corpus, and against Mistral's 103.1%. Both are more than twice the threshold this entry set for worth taking, so for those two families the question is answered and the answer is yes.

    It stays unscheduled anyway, and the reason is not the measurement. A real tokenizer per family is a dependency in packages that have none, and that is a distribution decision rather than a measurement one — the same shape as the editor extension above. What has changed is that it can no longer be deferred for want of evidence: the report now names the family and prints the measured error where one exists, and says nobody has measured it where one does not. Google, xAI and Moonshot are the three still unmeasured, each needs a key for its own counter, and each has its own named skip in the suite carrying the command to run. OpenAI left that list in 1.85.0, not because a key arrived but because its measurement was already committed here while this file and band.ts both said otherwise.

    The dependency question is settled, and the answer was a separate package. @trazum/tokenizer-openai ships in 1.85.0: the core still declares nothing, and somebody who wants exact OpenAI counts installs twenty-two megabytes on purpose.

  • Prompt library. Shipped in 1.7.0, and the reasoning here was wrong in an instructive way. This entry said storing prompts "would mean sending them to a server. Trazum's privacy story is that it never does" — and it read as a principle when it was really a conflation of two things. Rule 1 binds the optimiser: the CLI sends nothing, needs no account, and works with the network unplugged, and none of that changed. It never said nobody may run a service that stores what they chose to save to it. Left visible rather than deleted, because a roadmap that quietly removes the entries it went back on is a roadmap with no record of having gone back on anything.

  • Cost alerting. A bot that watches a prompt's monthly estimate and speaks up when it crosses a threshold, or when caching effectiveness drops. Attractive, and unscheduled for the same reason as the two above: it needs somewhere to run and something to remember — a service holding other teams' prompt metrics on a schedule. check --max-tokens in CI already covers the threshold case for anyone content to have the answer arrive on a pull request instead of in Slack, which is most of the value for none of the surface area.


View on GitHub