Everything Foreweave does
The whole surface, honestly scoped: what is here today, what you have to ask us to switch on, and nothing we haven't built. Jump to what you came for.
01 — Build the plan
A deterministic critical-path engine underneath four ways of looking at the same schedule. Every committed edit recomputes the full critical path in under half a second on a thousand-task plan; plans are supported to five thousand tasks, and anything past two thousand recomputes in the background with progress rather than freezing on you.
A hierarchical work breakdown you reorder by dragging, with roll-up to summaries, milestones, deadlines, and both manual and automatically scheduled tasks. All four dependency types — finish-to-start, start-to-start, finish-to-finish, start-to-finish — with signed lead and lag in working time, six constraint types, and working-day calendars per project and per person.
A dependency cycle is refused with the chain named. An over-constrained task, negative slack, or a missed deadline is surfaced in a banner that walks you to each one — the engine will not shuffle a date to make the arithmetic look tidy. What you see is what the constraints actually produce.
Gantt with a frozen task panel, sticky date header, drag-to-create, bar move, edge resize and live preview. Grid for fast keyboard entry and in-place date editing. Network — a layered PERT diagram with the critical path picked out. Timeline, a condensed executive band that auto-fits and carries only the tasks you put on it.
Multi-step undo and redo over every edit. An append-only history of who changed what and when, filterable, plus time-travel on a single task to see how its dates moved across the life of the plan. Two people editing at once cannot silently clobber each other — a stale edit is refused, not merged behind your back.
One task-details panel edits every attribute, its links and its assignments, keyboard-navigable throughout. Big plans stay responsive because only the rows on screen are rendered — a two-hundred-task plan mounts a couple of dozen rows, not the whole outline.
Light and dark palettes, each meeting WCAG 2.1 AA contrast, with visible keyboard focus on every control, status never carried by colour alone, and motion that honours your reduced-motion setting.
02 — Plan with the agent
Most scheduling assistants take a paragraph and hand back a plausible list. This one refuses to start there. It builds a picture of the project first, agrees an approach with you in writing, and only then puts tasks on a chart — into a draft scenario you review as a diff.
What kind of programme this is, who has to approve what, how complex it really is. Complexity is scored from a published rubric and always shown with the factors that drove it, so you can argue with the score instead of guessing at it. Every required approval maps to an actual gate in the plan.
A kickoff conversation that collects the minimum dataset it needs — and never re-asks something you already supplied. You can walk away and resume. Every answer keeps its provenance, and what stays unknown is written down as an open question rather than quietly assumed.
Before a single task exists, the agent proposes how it intends to plan: phases, gates, the shape of the schedule. You accept, amend, or reject it. Until you accept one, it cannot create tasks ad hoc — which is the difference between a collaborator and a random number generator.
Decisions and lessons persist per project across sessions, so the next conversation doesn't start from nothing. Memory never crosses a project or organization boundary. Promoting a lesson to your wider line of business takes an explicit human acceptance, and strips identifying detail on the way.
And when it does write
03 — Scenarios & Monte Carlo
A single finish date is a claim about a world where nothing varies. Foreweave lets you branch the plan, and lets you ask what the arithmetic actually says about when you will finish.
A what-if never touches your baseline. Compare two branches side by side — finish dates, and precisely which tasks enter or leave the critical path between them, which is usually the argument you were trying to make. Promote the winner in one atomic step, recorded in history.
Three-point estimates or named distributions, per task, per scenario. Five thousand iterations over a thousand-task plan run in under thirty seconds, vectorised across the whole iteration space rather than looped.
P10, P50, P80 and P90 with a finish histogram and an S-curve marked at the percentiles you commit on. A criticality index per task — the share of runs in which it actually sat on the critical path. Tornado sensitivity ranking what your finish date is genuinely hostage to.
Every run stores its seed and a frozen snapshot of the overrides and uncertainty it used, so a result from last quarter reproduces byte for byte — regardless of how many workers ran it. The Monte Carlo standard error is reported alongside the percentiles, so you can tell a converged answer from a noisy one.
04 — People & workload
Work equals duration times units, with task types that behave the way a planner expects when you change one of the three. Resources carry maximum units and their own working calendars.
Per task, per resource, and across the project, with a lane view showing each person's peak simultaneous demand. No end-of-month surprise about who was booked at 300%.
The leveller delays within available slack and reports exactly what it would move and why. It proposes rather than applies, and it runs only when asked — never a background process quietly rewriting your dates. It is reached through the API today; the in-app control to run it and review its proposal is still to come.
Budget at completion, planned value, earned value and actual cost, with schedule and cost performance indices, rolled up the breakdown structure.
05 — Baselines & actuals
Capture a named baseline whenever you commit to something. It renders as a ghost bar behind the live schedule, and the variance table sorts worst-slip-first in working days. Re-baselining keeps every prior baseline and shows up in history.
Record actual starts, finishes, percent complete and actual cost. Actuals take precedence over every other rule in the engine — a task that started on Tuesday started on Tuesday, whatever the constraint said. Percent transitions stamp dates for you rather than making you type them twice.
Set where "now" is for the plan, then sweep: the completed past is frozen and everything unfinished reschedules forward from the status date in one reviewable operation.
A recurring progress review surfaces what should have started, what should have finished, and what has been sitting at the same percentage for too long — so the update meeting has an agenda before it starts.
06 — Programs
Roll several project schedules into one program master. Members appear as read-only mirrors under their own summary task; you add cross-project links and program-only work on top. The interesting part is what happens when a member changes.
A member sends changes upward deliberately, as a net difference — edit something and revert it and nothing publishes at all. Until they publish, their work in progress is invisible to every master.
Changes that don't move the master's critical path or finish date apply automatically with an indicator. Anything that does queues for your decision, with a preview computed as a real scenario overlay — so what you accept is exactly what you were shown.
Reject a change and the master stays put, the difference is rendered as drift against the plan of record, and a persistent alert lands on the originating task naming the master that refused it. A later edit to the same task supersedes the decision and is judged afresh.
Actuals always apply, even when they move the critical path — you are told, you don't get a veto. A project can belong to several masters at once with independent verdicts, and masters can nest inside larger masters, with acceptance flowing up only on an explicit re-publish.
07 — Import & export
Tasks, links, constraints, deadlines, calendars, resources, assignments and actuals round-trip. Export the whole plan back out at any time through the API — there is no version of this product where your schedule is hostage to it. We read the XML export, not the binary .mpp, and say so up front.
Upload a CSV or XLSX — including the kind where the schedule is drawn as coloured cells, merged bars or drawing shapes rather than typed dates. Predecessor notation is parsed, dependency types and lags included. Files up to 25 MB and 50 sheets; anything large processes in the background with progress.
Extraction is presented for review with cell-level provenance on every item. The date-axis interpretation is confirmed before a single date is committed, low-confidence items need item-by-item acceptance, and nothing reaches a baseline without you accepting it. Re-uploading the same file gives the same answer.
Server-rendered Gantt and Timeline to PDF — vector, selectable text, paginated without dropping rows — plus PNG, and a task-table CSV. PNG is capped at 500 rows and tells you to use PDF or XML past that, rather than producing something unreadable.
08 — In preview
These two are finished and covered by tests, but they are dark by default at the platform level — one because it sends more of your data further than the rest of the product, the other because it reaches the open web. Neither turns on by accident, and we would rather list them honestly here than have you find them in a release note.
Ask for a steering-committee deck, a status report or a one-page brief and get a real .pptx, .docx or .pdf built from a slice of schedule you define. It is a read-only surface — there is no path from it back into your plan. Because the document-authoring path is not zero-retention, it needs a recorded organization-admin acknowledgement on top of the usual gates. The security page spells that out in full.
For genuinely complex programmes, the agent can research which approvals and certifications your domain actually demands — from the live web, with a source citation on every finding. It proposes the research and makes no external call until you accept it, confidential terms are stripped from the query first, and nothing is ever applied to your project profile without you selecting it.
09 — API, tenancy & security
Every organization is isolated, enforced twice: authorization checks tenant scope before role, and row-level security in the database backs it up and fails closed. Join through your company directory via OpenID Connect, or by single-use, expiring, revocable invitation.
An organization role governs membership; a per-project role — Owner, Editor, Viewer, plus a project administrator — governs every action, and the same person can hold different roles on different plans. Ownership transfers and project deletion are gated and audited.
A REST API with hashed, expiring bearer tokens and optimistic concurrency through schedule version tokens, so an integration cannot overwrite a change it never saw. The docs hub lists the endpoints.
Configuration fails closed, uploads and exports are bounded, XML is parsed with entity expansion disabled, dependencies are pinned by hash and audited every build. The full picture, including what we have chosen not to claim, is on the security page.
Seen enough
Import what you have or grow one from scratch. Free while in preview.