What is TaskTreker, and how is it different from a task tracker?
TaskTreker is a personal assistant for a product manager. It reads what already happened — meetings, documents, customer feedback, calendar, chat and email — and turns it into tasks, decisions, insights and facts, each linked to the source fragment. A task tracker stores tasks you entered yourself. TaskTreker also collects what you missed: an agreement made on a call, a request buried in a document, a deadline nobody wrote down.
Four layers of the assistant:
- Capture — the input: six channel types produce tasks, decisions, insights and facts with a link to the source fragment.
- Harness — the working layer: a prompt library grouped by task type, agent instructions, rituals, templates, the rules of your company and quality scoring.
- Assistant — the proactive layer: digest, missed list, suggested priority, assignee and delivery format, reminders.
- Check — the naive beginner test of your own service, direct feedback, and learning on your material.
The difference from a task tracker is the layer above it: what was missed, who should take it, and in what format. TaskTreker does not replace Jira or Linear as a system for bugs and sprints, and it does not keep a second queue of tasks.
How does TaskTreker find out what happened?
Capture is the input layer of TaskTreker. It connects to six channel types — meetings, documents, customer feedback, calendar, chat and email — and extracts typed nodes from what arrives: tasks, decisions, insights and facts. Every extracted node keeps a pointer to its origin, a timestamp, page, paragraph or thread, so you can open the original and check it. Extraction runs in the background and does not block your work.
What each channel gives:
- Meetings — a transcript or an uploaded recording: decisions, agreements, open questions, tasks.
- Documents — facts, assumptions and requirements, each with a link to the page or paragraph.
- Customer feedback — requests and complaints, bound to a feature so they stop dying in the inbox.
- Calendar — where the plan breaks: overloaded days, meetings without a decision, slots without work.
- Chat — decisions and questions that stayed inside a thread.
- Mail — requests, deadlines and promises that arrived by letter.
The channels are connected in waves: calendar, documents and meetings come first, then chat, mail and customer feedback. TaskTreker does not record a meeting live — processing starts when the channel has delivered the material.
What is the harness, and why does it matter?
The harness is the ready-made working layer around the model, and it saves you from assembling an agent yourself. It holds a prompt library grouped by task type, agent instructions, rituals for the day, the week and the month, templates for common outputs, the rules of your company format, and a quality bar for every prompt. The library ships ready and is maintained with TaskTreker.
What is inside:
- Prompt library — discovery, decision, handoff, retro, sizing, requirements, learning, digest; each prompt has an instruction, an input, an output and a quality criterion.
- Agent instructions — how the assistant treats your context, your rules and your language.
- Rituals — day, week and month; they run on a schedule and collect the material themselves.
- Templates — ready shapes for the results you produce most often, plus the rules of your company.
- Quality bar — every version of every prompt is scored on a set of examples.
TaskTreker does not leave the maintenance of these prompts to you: they are versioned, scored and rolled back inside TaskTreker.
How does the digest work, and when does it arrive?
The digest is the summary TaskTreker assembles instead of you collecting the morning yourself. It has three blocks: what changed, what is on fire, and what was missed. It arrives in the channel you choose — mail, chat or the web page — and it can be tied to a daily ritual so it comes on a schedule. A skipped ritual is recorded in the log instead of disappearing quietly.
What is in the digest and around it:
- Three blocks — what changed, what is on fire, what was missed.
- The channel — mail, chat or the web page, your choice.
- The schedule — tied to a daily ritual.
- Skipped rituals — visible in the log.
- Reminders — priority ones go into the digest, the rest arrive later, and any type can be switched off.
Reminders are built the same way: each one is tied to an event and a context, not to a fixed schedule of pressure. Priority items go into the digest, the rest arrive later, and any type of reminder can be switched off on its own. Silence is treated as a valid state — a day when nothing needs your attention is not a failure. TaskTreker does not send a notification for everything it noticed.
What does "missed" mean, and how does it work?
"Missed" is a separate list of things that were said but never became a task: an agreement made on a call, a request inside a document, a deadline mentioned in passing. TaskTreker keeps each entry with the quote, a link to the fragment of the source, why it matters and by when. It is the first place where you can see that the assistant caught more than a person did.
What each entry holds:
- What was said — the quote itself, not a paraphrase.
- Where — a link to the fragment: timestamp, page, paragraph or thread.
- Why it matters — the reasoning, tied to your context.
- By when — the deadline, if one was mentioned.
Confirming an entry turns it into a real task, and an entry left unconfirmed stays out of your context and out of the answers.
What is a sizing document, and how is it different from a TAM calculator?
A sizing document in TaskTreker is a calculation node inside the context graph, not a spreadsheet with cells. Every assumption is stored as an object: value, unit, range, source, author and date. Four families of calculation are available — top-down from TAM to SAM and SOM, bottom-up from units times price, feature ROI, and unit economics with CAC, LTV and payback.
The four families:
- Top-down — TAM to SAM to SOM: from the available market down to the reachable share.
- Bottom-up — units times price: how many units at what price.
- Feature ROI — the economics of one feature, on the same assumptions as the sizing.
- Unit economics — CAC, LTV and payback period.
The difference with a TAM calculator is in the links and the trace: change the price once and everything depending on it is recalculated, every version keeps the assumptions it was built on, and a value proposed by the model without a source is marked as an estimate and stays out of the result until you confirm it. TaskTreker does not produce ready-made market reports.
What is a handoff artifact, and which types exist?
A handoff artifact is a piece of material about a release, generated from your context instead of being written from scratch for every launch. TaskTreker prepares nine types of artifact, all built from the same context and differing only in audience and template: external release note, internal changelog, battlecard, outbound customer email, sales brief, sales FAQ, demo script, development task draft, and release announcement.
The nine types and who they are for:
- External release note — for users of the service: what changed, why, how to use it.
- Internal changelog — for the team: what changed, what it leads to, what to watch out for.
- Battlecard — for sales: differences, objections, and what must not be promised.
- Customer email — for clients: a personalised letter with the link and the value.
- Sales brief — for the sales team: who benefits, what to talk about and what to keep quiet about.
- Sales FAQ — for sales: typical client questions with answers linked to a source.
- Demo script — for sales and product marketing: a step-by-step scenario with the expected result.
- Development task draft — for engineers: a task or spec with a description and acceptance criteria.
- Release announcement — for external communication: the public announcement of the launch.
Every package passes a confirmation gate: publishing without a person is impossible. Each artifact keeps its versions, each assertion carries a source, and where the context lacks a positioning, a metric or a decision, the artifact shows an explicit gap instead of a plausible sentence. TaskTreker does not send anything to customers without approval, and the public changelog of your own service is not one of the nine types — the internal changelog is.
How do I connect Linear, Jira, GitHub, Notion and Slack?
Each integration is a separate connection with its own access rights, switched on from the integrations screen: you authorise the account, choose what TaskTreker may read and what may be written, and the channel starts delivering material. Twelve integrations are in scope for the first release: Linear, Jira, GitHub, Notion, Confluence, Slack, Google Docs and Drive, Figma, Amplitude or Mixpanel, HubSpot or amoCRM, mail, and your own MCP server.
The twelve and what each is for:
- Linear and Jira — your tasks, the status of a feature and of a release, and where draft tasks land.
- GitHub — pull requests and issues, the link between a release and the code.
- Notion and Confluence — your knowledge base and documents.
- Slack — delivery of artifacts, notifications, and decisions made in a channel.
- Google Docs and Drive — delivery of artifacts and external documents.
- Figma — mockups and screenshots of features.
- Amplitude or Mixpanel — product metrics for your assumption registry.
- HubSpot or amoCRM — client context and delivery into a deal card.
- Mail — sending artifacts and letters to clients.
- Your own MCP server — an agent with access to your context.
Any integration can be switched off, and the data can be exported without loss: TaskTreker does not require a migration to start working. The order in which the integrations are connected is still being decided and is announced with the release.
Does TaskTreker replace Jira or Linear?
No. TaskTreker is not a replacement for Jira or Linear as a system for bugs and sprints, and it keeps no second queue of the same tasks. You get one list where every row states where the task came from — Linear, Jira, your own task in the assistant, a finding from a meeting breakdown — and a transition to the original task. TaskTreker adds the layer above the tracker.
That layer is made of four things:
- What was missed, and where it was said.
- Who should take a task, and in what format.
- A priority proposed with its reasoning.
- A plan for the week when it does not fit.
The priority is proposed, not applied: it can be overridden by hand without an explanation, and nothing is changed in your tracker unless you confirm it. Where the tasks physically live is a product decision that has not been closed yet, and the behaviour is the same in every case.
Does TaskTreker replace Notion and ChatGPT?
No, and the two do different things. Notion is where you write and keep documents; TaskTreker reads the channels you connect, extracts typed nodes, links them to each other and answers questions about them with citations. ChatGPT answers from a model; TaskTreker answers from your stored context — strategy, goals, hypotheses, assumptions, decisions, artifacts — and says so plainly when the context holds no answer at all.
The difference in one line each:
- Notion — a place to write and keep documents; TaskTreker — a place where the reasoning behind the work is linked and searchable.
- ChatGPT — answers from a model; TaskTreker — answers from your stored context, with citations, or says that there is no answer.
TaskTreker is not a separate chat: context search and answers live inside the TaskTreker interface, and the MCP server is a second surface for agents such as Claude Code or Cursor. It is also not a note-taking tool — what it keeps is the reasoning behind the work, decisions with their alternatives and assumptions with their sources, and every answer about them points back to the source.
Where is the data stored?
The data lives in TaskTreker's own database, in an isolated space per company, and it stays exportable. TaskTreker does not ship your data to an outside analytics service and does not build a separate BI layer over analytics — it connects to the analytics you already use.
What you can do with the data:
- Import and export — Markdown, JSON and CSV, without losing the structure.
- Isolation — a separate space per company; another tenant is not reachable.
- Analytics — you connect Amplitude, Mixpanel or another system, not a separate BI.
- Deletion — export and removal on request, with stated retention periods.
- Legal regime — GDPR as the base, 152-FZ as an additional regime for the Russian segment.
Where the data physically sits and where the model processing runs is still an open decision, so no location is published. The base legal regime is GDPR, with 152-FZ as an additional regime for the Russian segment rather than the base one. Data can be exported and deleted on request, and the retention periods are stated rather than left implicit. TaskTreker does not create its own copy of your materials in an outside system, and it does not keep a hidden copy after a deletion request.
Who can see our team's data?
Only the people of your own company. Isolation between companies is enforced at the database level, not by convention: every request runs in a transaction scoped to your company, and access to a resource belonging to another company is refused without revealing whether that resource exists. Inside the company, rights are limited by role and every action is written to the audit log.
Roles inside a company:
- Product manager — the full picture: context, tasks, artifacts, sizing.
- Product marketing — the materials and their tone, without the internal drafts.
- Sales — the released materials and the answers, without the internal assumptions.
- Observer — read-only.
The audit log records who created, changed, confirmed, published or deleted something, and when; every request to a model and every call through the MCP server is written there too. TaskTreker does not give an agent or an integration access beyond what it was authorised for.
Which data is passed to the AI?
Only what you allow, and only with explicit consent: TaskTreker does not send data to a model on its own. Every connected channel carries its own rights — what is read, what is not, and what may leave the workspace. Sensitive material can be marked, and sending it can be switched off while local processing remains.
The rules that apply:
- Rights per channel — what is read, what is not, what is sent to the model.
- Consent — explicit and revocable; without it the channel is not connected.
- Sensitive material — a mark, a switch that stops sending, and local processing instead.
- Untrusted input — instructions inside your content stay data, not orders.
- Audit — every request to a model is written to the log.
Your content is treated as untrusted input: instructions inside a document, a message or a recording are data, not orders, and they cannot make the model reveal anything or change its behaviour. And the model never invents numbers — a market value without a source is marked as an estimate, needs confirmation, and stays out of a published artifact until a person signs off.
Does TaskTreker work in Russian and English?
Yes. TaskTreker is English-first: English is the language of the interface, of the harness, of the prompts, of the rituals and of the digests, and Russian is supported as a second language. The prompt library and the evaluation sets are written in English as the source language, and the evaluation runs on every supported language, so the quality of the harness does not depend on the language of the interface.
What is localised:
- The interface — English first, Russian supported.
- The harness — prompts, instructions, rituals and evaluation sets, written in English as the source.
- Dates, currency, numbers, time and formats — by the settings of your company, not by the interface language.
- Artifacts — the internal changelog is part of the package; translating every artifact is still an open question.
The library itself is written in English and translated from there: translating the source would put the quality bar of the harness onto the quality of the translation. TaskTreker does not translate every artifact into every language in the first release, and the localisation of the exported artifacts is still an open question.
How much does TaskTreker cost?
Four tariffs: Free, 299 ₽, 499 ₽ and 999 ₽ per month for the Pro plan. The price is per company, not per seat — a developer who occasionally looks into TaskTreker is not billed separately, and the company pays. The model is freemium: you start on the free tariff and move to a paid one as the value grows. The trial period is 14 days.
How the payment works:
- Free — a self-serve entry, without approval.
- 299 ₽, 499 ₽ and 999 ₽ — the paid tariffs, the last one is the Pro plan.
- Flat price — per company, not per seat.
- Freemium — you start free and move to a paid tariff as the value grows.
- Trial — 14 days.
- Currency — rubles.
What is included in each tariff and what the limits are is still an open decision, and it will be announced before the public beta. The payment path is not connected yet, so the free tariff is what is available today.
How do I cancel the subscription and delete the data?
Export and deletion on request are stated requirements of TaskTreker, together with the retention periods and the order of deletion — nothing is removed on the initiative of the service. The same applies to a connected source: you switch it off, nothing new is read from it, and what was extracted earlier stays visible and exportable. The current state is worth stating plainly: the payment path is not connected yet.
How it works:
- A request — export and deletion are yours to ask for, not the service's initiative.
- Stated periods — retention periods and the order of operations are published.
- A switched-off source — nothing new is read; what was extracted stays exportable.
- No hidden copy — nothing is kept elsewhere after a deletion request.
So today there is no paid subscription to cancel: the free tariff is what is available, and the cancellation flow arrives together with the payment path. Deletion is not something TaskTreker does quietly on its own — the request goes through the same procedure as the export, with the retention periods and the order of operations stated in advance.
What happens if I connect a source with sensitive data?
The channel is connected with explicit rights, and sensitive material is marked and handled separately. You decide what TaskTreker may read from that source and what may be sent to a model; for sensitive material, sending can be switched off while local processing and the fragment-only mode remain. Every request to a model is audited.
What happens with a sensitive source:
- Rights of the channel — what is read, what is not, what is sent to the model.
- A sensitive mark — such material is shown separately from the rest.
- A switch that stops sending — local processing and the fragment-only mode remain.
- Consent to record — explicit and revocable before any processing starts.
- An audit trail — what was read, by whom and when.
For the meetings channel there is a hard precondition: an explicit, revocable consent to record, given at the channel or at the meeting level before any processing starts. Without that consent the meeting is not processed at all — not processed with a caveat. TaskTreker does not record a meeting live from the microphone in the first release: processing starts when the channel has delivered the material.
How do I get the answer to "why did we do this"?
Ask in free form — for example, why the team moved away from a free plan — and TaskTreker assembles the answer from the context layer: strategy, goals, initiatives, hypotheses, assumptions, decisions, features and artifacts, ranked by relevance and recency. The answer shows the chain, assumption to hypothesis to decision to consequence, and every factual statement carries a citation to a specific node with its date and its freshness.
Three parts are always present:
- The reason — the assumption that was being tested.
- The decision — what was chosen.
- The context of the moment — the price, the volume and the state of the market at that time.
Without the third part, an answer becomes misleading a year later. TaskTreker does not answer from general knowledge when the context holds nothing: there it says so and offers to create a decision record.
What happens if the context holds no answer?
TaskTreker says so plainly and offers to create a decision record instead of assembling a plausible explanation. If no decision is recorded on the question, the answer states that directly and the sources open, so you can see what was written at the time. The same rule governs market numbers.
The rules when there is no answer:
- No decision record — the answer says so in plain words.
- A market figure without a source — marked as an estimate.
- An unconfirmed estimate — does not reach a published artifact.
- No answer at all — a proposal to record the decision.
A value proposed by the model without a source is marked as an estimate, requires confirmation, and cannot be used in a published artifact until a person signs off. An empty answer is a valid result, not a defect: TaskTreker does not fill the gap with something that sounds right.
How does TaskTreker help me manage tasks and time?
You get one list of tasks where every row states its source, and for each task TaskTreker proposes a priority with its reasoning: the deadline, the dependency, the goal. The proposal can be overridden by hand without explaining anything, and nothing in your tracker changes automatically. When the week does not fit, TaskTreker suggests what to postpone, delegate or narrow, and a weekly breakdown shows how the time went.
What the assistant does with tasks and time:
- One list — Linear, Jira, your own tasks in the assistant, and findings from meeting breakdowns.
- Suggested priority — with the reasoning; you override it by hand without explaining.
- A series of postponements — read back as a blocker, an overload or a mismatch with the plan.
- An overloaded week — what to postpone, delegate or narrow, with the consequence of each option shown before you choose.
- A weekly breakdown — how much time went into work that moves your roadmap forward, and how much did not.
TaskTreker does not apply a priority, a postponement or a delegation on its own: proposals stay proposals until you act.
How does TaskTreker deliver artifacts?
Seven channels: inside TaskTreker, email, Slack or Teams, Notion or Confluence, a CRM such as HubSpot, amoCRM or Salesforce, Google Docs, and a web page by link. You pick the channel when you deliver the package — an unavailable channel does not block the others — and each delivery is written to the event log with the channel and the result.
The seven channels:
- Inside TaskTreker — read and confirm without leaving the interface.
- Email — a letter with a link to the package.
- Slack or Teams — delivery into a working chat.
- Notion or Confluence — a page in your knowledge base.
- CRM — the deal card.
- Google Docs — a document with the materials.
- Web page by link — access without an account.
Development task drafts are prepared in one of four formats: feature, spec, epic, ticket. A ticket or a spec arrives with a description and explicit acceptance criteria, and a feature or an epic carries them at its own level. TaskTreker does not publish anything without confirmation and does not send anything to customers on its own.
How is the quality of the harness measured?
Each prompt in the harness has a version and a score on a set of examples with expected results. A change is accepted only after it has been run against that set, and TaskTreker shows the result before and after. A rollback returns the previous version without losing history, and a quality dashboard shows where the assistant made a mistake, where you rewrote the output and what was missed.
How the measurement works:
- A version — every prompt has its own number and its own score.
- A set of examples — reference cases with the expected result.
- Before and after — what the edit changed, side by side.
- Rollback — one click, and the history stays.
- A quality dashboard — where the assistant was wrong, where you rewrote the output, and what it skipped.
The signals in that dashboard come from real runs and feed the improvement of the library; they are not a complaint about the person who rewrote the output.
Does TaskTreker take the format of my company into account?
Yes, and this is the company profile: guidelines, tone, terminology, templates and prohibitions. TaskTreker checks its own results against those rules, and when something does not match, it shows which rule was violated and where that rule is written down, so you fix the cause rather than the symptom. Your own templates can be imported, and a company template takes priority over the library one.
What the company profile holds:
- Guidelines — what is allowed and what is not in the material.
- Tone and terminology — how the company sounds and what it calls things.
- Templates — the shapes you use most often.
- Prohibitions — what must not reach the result.
- A check of the result — on a mismatch, the rule and the place it is written down are shown.
Adaptation of the harness to your working style works the same way: it is offered with an explanation of what will change and why, and applied only after you agree; refusing needs no explanation. TaskTreker does not rewrite your rules on its own and does not replace them with the library ones — the profile describes your company, and the library stays a starting point.
What is the "dumb user" test?
It is the check layer of TaskTreker: the assistant walks through your own service the way a naive newcomer would — without knowledge of the internal structure, allowed to ask a silly question, without hints from the documentation. The result is a map of where the newcomer got stuck, with reproducible steps, and the feedback is direct: what broke, how much it interferes, what to change.
What the test does:
- The scenario — a naive newcomer without knowledge of the internal structure.
- A map of sticking points — where the run stopped, what was not understood, which path was not found.
- Direct feedback — what broke, how much it interferes, what to change.
- The backlog — an accepted finding becomes a task with a priority and a reproduction.
- Boundaries — the test writes nothing to your systems and changes no data.
A finding you accept becomes a task in the backlog with a priority, a link to the reproduction and an impact assessment. The test stays inside your own service: it does not reach internal systems, does not write to them and does not change data — it runs on the demo space or on a stand you provide, and a run on a client stand needs your explicit permission. TaskTreker does not soften the wording: "generally fine" is not an accepted form of feedback.
How does TaskTreker help me learn something new?
You choose a method — sizing, discovery, writing a PRD, prioritisation — and TaskTreker works through it on your own material first, explaining the general principle only afterwards. The walkthrough relies on real data from your company and says so plainly when the data is not enough, proposing what to collect. Afterwards it checks your understanding with a short question and works through the mistakes you made.
How the learning runs:
- Your example first — the walkthrough is built on your assumptions and your data.
- The general principle afterwards — the rule of the method, not a lecture.
- A check of understanding — a short question and a breakdown of the errors.
- Repetition of weak topics — a topic with mistakes comes back in the next checks, and your progress is visible.
Learning does not change your context: anything you want to keep is recorded as a decision and enters the context through the usual confirmation path.
How do I start with TaskTreker?
Start with three steps: register with email and password without approval, connect two or three channels, and get the first digest. The first screen asks which channels to connect rather than for a description of your company — forms are not the entry point. There is a demo space that runs on demo data, so you can walk through the scenario first.
The three steps:
- Registration — email and password, no approval needed.
- Connecting channels — two or three are enough; all six are not required.
- The first digest — the same day, from the channels you connected.
Connecting all six channels is not required: two or three are enough to start. The empty start is not left looking broken — you see what is already connected, what will arrive next, and what happens once there is enough material. TaskTreker does not ask you to describe your company before it shows you anything.
What does the MCP server of TaskTreker give me?
TaskTreker exposes an MCP server, so an agent you already use — Claude Code, Cursor — can query your context. Four tools are available: list the context nodes, get one node by identifier, list the tasks with their source and the reasoning behind the priority, and ask a question about the company in natural language.
The four tools:
- list_context_nodes — context nodes: strategies, goals, hypotheses, assumptions, decisions.
- get_context_node — the full text of one node by its identifier.
- list_tasks — open tasks with their source, deadline and priority reasoning.
- ask_company — a question about the company in natural language, answered with citations.
Answers are assembled from stored context only and come with citations to the nodes they were built from. By default the server works in read-only mode, every call is written to the audit log, and an agent sees only the company it was authorised for. The MCP server does not turn TaskTreker into a task tracker and does not change what TaskTreker is: it is a second surface for the same context, next to the search inside the interface. Writes are possible only where they are explicitly allowed, and they are logged too.
Does TaskTreker work on a phone?
The web version of TaskTreker is adaptive, so the basic scenarios work on a phone: reading the context, reading and confirming an artifact package, asking questions about your context, and looking at the dashboard of gaps. Writing and editing the context from a phone is not required for the basic flow. Whether to go further with an installable web app or a native application is still being decided.
What works from a phone:
- Reading the context — goals, decisions, assumptions.
- Reading and confirming a package of artifacts.
- Asking questions about your own context.
- The dashboard of gaps.
That decision is not announced, so nothing further is promised on this point. What is fixed for the first release is the adaptive web version and the basic scenarios on it: viewing the context, reading and confirming a package of artifacts, asking questions about your context, and the dashboard of gaps.
What does TaskTreker not do?
The boundaries are stated up front rather than discovered later. TaskTreker does not replace Jira or Linear as a system for bugs and sprints, does not become a corporate messenger and does not duplicate your mail. It does not write code for you and does not build a separate BI layer over analytics.
What TaskTreker does not do:
- It does not replace Jira and Linear as a system for bugs and sprints.
- It does not become a corporate messenger and does not duplicate mail.
- It does not write code and does not run deployments.
- It does not build a separate BI over your analytics.
- It does not include a visual diagram editor.
- It does not invent market numbers — it calculates only from the assumptions you entered.
- It does not publish anything without confirmation by a person.
- It does not keep its own queue of tasks — the tasks stay yours.
TaskTreker also does not produce ready-made market reports or statistics pulled from external sources in real time: every calculation uses the assumptions you entered, and each of them can be traced back to its source.
How do I move my existing context into TaskTreker?
Through the universal path: import in Markdown, JSON or CSV, without rewriting your materials. Strategy documents, decisions and notes come in as they are, with authorship and dates preserved where the source contains them, and imported material is marked as not verified until someone takes ownership of it. The import is idempotent — running it twice does not create duplicates.
What the import guarantees:
- Formats — Markdown, JSON and CSV.
- Authorship and dates — preserved where the source has them.
- Status — imported material is marked as not verified.
- Idempotence — a second import does not create duplicates.
- No migration — reading through a connector does not require moving your work either.
If your stack has no dedicated integration, the import path covers it while the connector is being built: the universal path is not a fallback for a rare stack, it is the way in. TaskTreker does not ask you to move your work into a new tool before it shows you anything.
What works today in TaskTreker, and what is still being built?
Already working: capture from connected channels with the source kept for every extracted item, the missed list, the context layer that answers why a decision was made, the harness with nine prompts in two languages, and twelve connectors. In development: automatic polling of connected sources, the tariff breakdown, and the payment path. Not there yet: a demo without registration and in Russian, release version numbers and dates.
Where TaskTreker stands:
- Already working — capture with the source of every extracted item, the missed list, the context with an answer to why a decision was made, the harness with nine prompts in two languages, twelve connectors, import and export in Markdown, JSON and CSV, isolation between companies, the audit log.
- In development — automatic polling of connected sources, the breakdown of features and limits per tariff, the payment path.
- Not there yet — a demo without registration and in Russian, release version numbers and dates.
The release process itself has not been defined, so no version numbers or dates are published: a record appears only after a person has checked it. The public beta is being prepared with a self-serve entry and a demo space on demo data.