Astro

Astro is Orgonaut's in-app assistant. It is tenant-wide — a single unified assistant per tenant and user that works across Live baseline, scenarios, and snapshots. Astro can read from any context and write to active scenarios. It includes governed scenario-only change flows for actor records, actor-adjacent placements and compensation, team and department changes, org-unit structure edits, positions, scenario-linked initiatives, and scenario lifecycle drafting. Tenant settings remain readable, but their mutations are still human-owned. It has two surfaces:

a primary Astro workspace page accessible via the Astro link in the sidebar
a secondary floating overlay for quick questions while you browse

Use the workspace when you want a longer analysis session. Use the overlay when you want to ask about the page you are already on without leaving it.

Astro also includes a docs-grounded knowledge capability for product-help questions. When you ask how-to, troubleshooting, workflow, API, CLI, or MCP questions, Astro grounds the answer in indexed Orgonaut documentation instead of guessing from generic model knowledge.

If you want copy-ready examples, use the Astro prompt cheat sheet. It is grounded in the seeded demo tenant and walks through live analysis, scenario drafting, AI-first planning changes, and promotion review prompts.

Where to find it

Click Astro in the sidebar to open the workspace. The link is scenario-aware — when you are viewing a scenario, it opens /scenarios/{slug}/ai; for a snapshot, /snapshots/{slug}/ai; and for Live baseline, /ai.
On non-Astro pages, a floating overlay is available for quick questions without leaving the current page.
On mobile, the overflow menu in the header also includes an Astro link.
A colour-coded badge at the top of both the workspace and overlay shows the current context: emerald for Live baseline, amber for scenarios, and zinc for snapshots.
In the dedicated workspace, the composer stays pinned to the bottom of the viewport while the main thread area scrolls independently.

If your role does not include assistant access, the AI button is hidden.

What it can do in v1

Astro is limited to Orgonaut's structured organisational data and governed scenario planning flows. It can help with questions such as:

what changed recently in a scenario
what scenario tasks are running, whether a scenario has been promoted, and which snapshots or lineage events are tied to it
which departments exist in any context — Live baseline, any scenario, or any snapshot
how a team is staffed today
what a simple read-only move estimate would do to team or department headcount, including human, agent, or robot-specific moves such as moving three people from Engineering to Product
what a multi-phase restructure plan would look like for broad goals such as AI-first operating model changes, including AI-generated objectives, assumptions, constraints, phased actions, estimated deltas, risks, dependency graph, and an execution handoff to governed scenario proposal bundles
inside a mutable scenario, turn a broad restructure objective into a phased governed program checkpoint with per-phase preview metrics, dependency checks, rollback points, and queued confirmation gates
which open roles or budgeted positions exist in any context
which initiatives are linked to a scenario and what their baseline and target scenarios are
what a team's recent sprint velocity looks like
who sits directly in a department today and which memberships are primary
where a department's cost concentration sits
where an actor sits, what else they are allocated to, and what their current compensation slice looks like when your permissions allow it
which tenant members, invitations, API tokens, MCP connections, and AI model pricing rows exist in the current tenant settings
what plan the tenant is on, how many actors count against the plan limit, and how many Astro turns have been used this month
whether BambooHR is connected, what recent sync logs say, which import runs are still processing, and whether exports are available
how to promote scenarios, troubleshoot imports, connect integrations, or find supported workflows from grounded Orgonaut docs
where to find API, CLI, and MCP reference guidance when those artifacts are indexed for your environment
what to review next on the current page
stage a new person, agent, or robot in a mutable scenario after explicit confirmation
stage a governed update or archive for an existing scenario-scoped actor after explicit confirmation
stage a governed placement create, update, or end for a scenario-scoped actor after explicit confirmation
stage a governed compensation create, update, or delete for a scenario-scoped actor after explicit confirmation
draft a governed bulk compensation proposal in a mutable scenario, including selector-based previews, before/after totals, sampled record changes, and explicit confirmation before any writes
draft a governed placement redistribution proposal in a mutable scenario, including before/after team coverage, variance reduction previews, flagged risks, and explicit confirmation before any writes
stage a governed agent cost-event create for a scenario-scoped agent after explicit confirmation
stage a governed team create, update, or delete inside a mutable scenario after explicit confirmation
stage a governed department create, update, or delete inside a mutable scenario after explicit confirmation
stage governed org-unit create, update, move, merge, split, delete, and restore changes inside a mutable scenario after explicit confirmation
stage governed position create, update, or delete changes inside a mutable scenario after explicit confirmation
stage governed initiative create, update, or delete changes that stay linked to the current scenario after explicit confirmation
create a new planning scenario draft, update scenario metadata, or submit a mutable planning scenario for human review after explicit confirmation when the current tenant plan includes scenario drafting
trigger PNG export on Charts and PNG, SVG, or CSV export on Matrix from the current visualization surface
send compatible scenario-scoped visualization modeling requests into the current chart or matrix change tray instead of executing them directly

Astro now presents its discovery surface in broader user-facing groups:

Explore organisation
Inspect teams and people
Analyse metrics
Work with scenarios
Inspect settings and operations
Make organisational changes

Behind the scenes, these discovery groups still map back to the shared Astro/MCP capability registry so the assistant, docs, and machine-readable metadata stay aligned.

Capability discovery

Astro can describe its own capability surface in the current tenant and context.

Ask What can Astro do? to get a grounded summary of capabilities that are available now, limited in the current context, or still coming soon.
Ask Can you do X here? to have Astro explain whether that workflow is supported for your current role and scope.
Ask Show me example prompts to get context-aware suggestions instead of guessing what Astro might understand.

Capability discovery is permission-aware and context-aware:

Astro does not claim a capability that your current role cannot use.
Astro distinguishes between read, proposal, and apply style workflows.
Broad help prompts such as What can you do?, What can Astro do here?, and Help me get started now use a direct server-side help path in both Live and scenario contexts, so Astro answers those quickly instead of refusing them in Live or expanding into a long recommendation frame.
Those broad help answers are intentionally minimal. Astro answers the question and suggests a few example prompts inline, but it usually does not add separate Why, Assumptions, Sources, How Astro worked, or follow-up action cards unless the turn actually needs them.
Astro can now stage governed actor and actor-adjacent scenario changes inside a mutable scenario.
Scenario lifecycle drafting is now available for create, update, and submit-for-review flows, while promotion and snapshot restore remain human-only.
In a mutable scenario, the capability browser now exposes real proposal cards for actor records, staffing changes, structure changes, and scenario lifecycle drafting instead of placeholder coming-soon entries.
Proposal capability cards now use a shared workflow pattern: Astro drafts a reviewable proposal first, shows the affected records and fields, and waits for explicit confirmation before anything executes.
For larger restructures, Astro can also draft bundled proposals that group multiple atomic changes into one reviewed plan with a single confirmation gate.
In a mutable scenario, Astro can also turn a broad restructure goal into a phased program. Astro now shows that program as one active checkpoint plus a compact timeline, keeps review-only checkpoints separate from executable bundle phases, and only drafts governed bundles when the active phase is concrete enough to execute safely.
When you reference an existing parent unit by name, including phrases like the new AI department, Astro now verifies that name against the current scenario before staging the change. If it still cannot resolve the parent cleanly, it asks for clarification instead of guessing an id.
The Astro workspace now includes a What Astro can do browser that presents one capability card at a time in a carousel, keeps unavailable items behind an explicit reveal, and lets you drop suggested prompts straight into the composer.
Empty-state prompt chips in both the workspace and overlay are now generated from the same capability catalog Astro uses for self-description, and Astro ranks those examples with both current-scope fit and recent prompt uptake signals.
When Astro cannot complete a request in the current scope, it now offers nearby supported prompts under Try instead rather than leaving recovery to trial-and-error.
Capability cards can now show multiple reasons when that matters, so snapshot, archived-scenario, and tenant-feature limits are explained more precisely instead of collapsing into one generic message.

When you ask about people, agents, or robots, Astro routes those requests through the shared actor domain.

person and people map to human actors
agent and robot keep their actor-kind filters
Astro does not treat them as separate entity families behind the scenes

Astro is tenant-wide. A single unified conversation persists per tenant and user, regardless of which page, baseline, scenario, or snapshot you are viewing. You can navigate freely between baseline, scenarios, and snapshots without losing your conversation thread.

Astro can read from any context automatically, including:

Live baseline
any scenario (draft, active, or archived)
any snapshot
the focused entity when you are on an actor, team, department, or scenario page and use the overlay
as of date when the page includes one

When Astro is answering from a historical page scope with an as of date, it carries that exact date through both the suggested prompt chips and the final answer text so historical reads are not described as “today”.

Astro can write only to active (mutable) scenarios. It cannot write to Live baseline, archived scenarios, or snapshots. Tenant-level settings are readable from any scope, but their mutations are still human-owned.

The dedicated Astro workspace anchors itself at the company root org unit for the current page context, rather than pinning the current actor, team, or department unless a future surface deliberately deep-links that context.

How answers are shown

Astro answers are structured so you can verify them quickly.

your question and Astro's reply use a denser thread layout so long sessions stay easy to scan in both the workspace and the overlay
markdown answers now use richer typography for headings, lists, inline code, and links, and wide tables scroll cleanly on smaller screens
when you send a question, the composer clears immediately and the thread scrolls to show your new question plus live runtime steps at the bottom while Astro finishes the turn
auto-scroll follows new replies only while you are already at the bottom of the thread; if you scroll up to review earlier messages, new replies arrive without disrupting your scroll position
when you scroll up in the thread, a floating Jump to bottom pill appears at the bottom of the chat; if a new reply arrives while scrolled up, the label changes to New reply so you know fresh content is waiting
clicking the pill or scrolling back to the bottom dismisses it and re-enables auto-scroll
once a reply has finished, the final scroll prefers the full question-plus-answer block when it fits, otherwise it settles at the start of the new answer instead of the bottom of the reply card

Each Astro reply includes:

Answer
Why when Astro needs to explain grounded reasoning or a constrained answer
Sources when Astro is grounding the reply in current records or indexed docs
Next useful action, now trimmed to one primary follow-up prompt plus one primary navigation action by default, with secondary options behind More options, for turns that actually benefit from an immediate next step
explicit navigation intent labels such as Open settings, Review changes, View activity, and Open guide so you can tell whether Astro is about to navigate or keep composing in chat
live pending-step telemetry while the turn is running, grounded in Astro's actual route, tool, and fallback events
an expandable How Astro worked panel in the details menu when Astro has stored runtime telemetry for the completed turn and the reply needs that extra inspection layer
quick actions to copy the reply or rate it as helpful or unhelpful

Straight capability/help replies are intentionally quieter than analysis or governed-change replies. If you ask a broad question like What can you do?, Astro now defaults to a concise answer plus a few inline examples instead of surrounding that answer with generic rationale, telemetry, and navigation chrome.

That telemetry is grounded in the real runtime, not canned copy. While Astro is still working, the pending card can now show the current route decision, active tool step, and fallback progress. Once the turn completes, the reply details can show the selected read or action path, the planner or runtime agents Astro used, per-agent output summaries, tool calls, retries, failovers, and the clarification reason when Astro had to stop for more input or permission limits.

When Astro cannot fully complete a request, the reply now also carries:

explicit Assumptions so you can see what Astro is holding constant
a focused Best next step question instead of only a flat limitation message
prompt-ready follow-up chips you can send back immediately to keep the conversation moving
clarification candidates with separate Use as prompt and Open record controls so you can either keep the conversation going or jump straight to the exact record Astro found

For product-help turns, Astro also tracks:

whether the answer stayed in structured data only, used docs only, or combined both (structured_only, knowledge_only, hybrid)
how confident the docs match was (high, medium, or low)
whether coverage looked complete or partial
explicit documentation gaps when Astro could not verify the whole workflow from indexed sources

When Astro drafts a governed scenario change, the answer card also includes:

a clear Proposal label so proposal turns do not read like plain analysis
the current proposal status, such as awaiting confirmation or already executed
the proposal state and operation
a reviewable preview of the staged change
an explicit confirmation button before Astro executes anything
bundle metadata when applicable, including strategy, item count, and per-item status details
for supported bulk proposals, typed bulk metadata such as affected counts, preview totals or coverage variance, warnings, and rollback mode when confirmation fails

On chart and matrix surfaces, Astro can also return:

a direct export action for the current visualization
an Add to change tray action when the request maps cleanly onto the existing scenario modeling tray

Substantive answers include source links back into Orgonaut so you can inspect the referenced team, department, actor, or scenario directly. Those source citations now use standard browser navigation so clicking them opens the linked page reliably. Knowledge answers also cite the relevant documentation page and heading when Astro used indexed docs. If Astro relied on indexed internal knowledge that is not exposed in the browser, the source list shows it as From the Orgonaut Knowledge Base instead of presenting a dead link.

The structured assistant payload also includes stable internal capability_ids. These IDs are used for roadmap tracking, provenance review, and future Astro/MCP parity work.

The same governed scenario action contract is now available over remote MCP through scenario.action.plan and scenario.action.confirm, while visualization exports and tray-bridging remain in-product only.

Bundled proposal execution currently uses a transactional rollback strategy:

Astro stages each bundle item in order for review.
Confirming the bundle executes all items inside one governed transaction in the current mutable scenario.
If one item fails validation or authorization, Astro rolls back prior writes from that confirmation attempt and reports rollback metadata with the failed item details.

Astro renders styled markdown inside the main answer body, including links, lists, headings, inline code, code blocks, and horizontally scrollable tables.

Astro answers now also carry stronger read-only grounding metadata behind the scenes:

sources are normalized into stable citations with scope-aware labels such as current scope versus related scenario
source summaries record how many cited records were linked and how many were inside the current scope
next useful actions are deterministic, route-backed follow-ups for the exact read surface Astro used, including positions, initiatives, scenario history, settings pages, integrations, and imports when those surfaces are available for the current tenant plan and your permissions
each answer includes a follow-up state so the UI and evals can distinguish between review, clarification, and limited-answer flows
docs-grounded turns record knowledge mode, confidence, coverage, and retrieval diagnostics so low-confidence answers can be reviewed explicitly
Astro now performs a server-side pre-resolution pass before each read turn. It extracts likely entity mentions from your question and injects ranked pre_resolved_entities hints (with confidence) into the turn context before model reasoning starts.
When pre-resolution confidence is low, Astro treats those hints as provisional and verifies with read tools before making substantive claims.
Entity resolution now uses ranked fuzzy matching with alias expansion (for example Eng → Engineering, SRE → Site Reliability Engineering) and returns confidence-scored candidates with match explanations.
When Astro cannot resolve cleanly on the first attempt, it automatically retries in a bounded sequence (broader type scope, alias-expanded query, top-candidate synthesis) before returning an ambiguity or limitation response.
Astro now scores question complexity (simple lookup, cross-entity synthesis, org-transformation) and routes complex read turns to a higher-quality read runtime policy. Simple lookups remain on the lower-latency mini policy.
When a question mixes workflow guidance with current tenant state, Astro combines both and keeps the answer split into Documented guidance and Current tenant data so you can see what comes from docs versus live records.
For broad transformation prompts, Astro can call a dedicated AI-backed restructure planner tool that returns a read-only phased plan with scenario-aware context, explicit source citations on every estimated metric, and a governed execution-handoff block for the next proposal step.
That restructure planner now includes built-in AI transformation playbooks as Option A/B/C: augmentation-first, automation-pod, and selective role redesign.
Astro now uses the planner model to assess and recommend the best-fit playbook (recommended_option_id) for the goal, with a short rationale.
Each playbook includes a trade-off table and a recommended execution sequence where every step references current org data tools plus source citations so you can verify the recommendation path before drafting scenario bundles.
The execution handoff now also includes per-option apply prompts that you can send directly back to Astro to draft governed scenario proposal bundles for the selected playbook (still confirmation-gated before any writes).

Choosing the right Astro workflow

Astro is most predictable when the prompt shape matches the workflow you actually want:

Use a direct question when you want one grounded answer about a team, department, actor, scenario, import, or settings surface. Astro keeps straightforward lookups on the faster read path.
Use a read-only restructure plan when the goal spans multiple teams or an org-wide change program. Astro routes those broader prompts through a higher-quality analysis path, then returns phased recommendations, Option A/B/C playbooks, a recommended option, and cited assumptions, risks, and deltas.
Use a phased scenario program when you are already in a mutable scenario and want Astro to decompose a broad restructure objective into sequential governed checkpoints. Astro drafts the program first, keeps review-only checkpoints out of the write path, and waits for confirmation before each executable phase bundle.
Use a bundled proposal prompt when you already know the coordinated scenario changes you want reviewed together. Astro drafts the grouped scenario proposal first, shows the bundle items, and still waits for one explicit confirmation before anything executes.

That bundled-proposal path now also covers supported governed bulk requests such as “increase all salaries in Engineering by 10% from 2026-07-01” or “redistribute all Scrum Masters evenly across Alpha, Beta, and Gamma.” Astro resolves the target set server-side, shows the preview, and only writes after you confirm the full proposal.

A good pattern for larger changes is: ask for the read-only plan first, review the recommended playbook, then send the plan's apply prompt or ask Astro to draft a bundled proposal from that option. Broad planning prompts can take a little longer than simple lookups because Astro verifies more entities, sources, and scenario constraints before answering.

Scenario sandbox guardrails

Astro now makes its change guardrails explicit on proposal turns so you can see what it is allowed to do before you confirm anything.

In Live baseline and snapshots, Astro stays in Hard rails. It can read, explain, and help you start a planning scenario, but it will not mutate baseline data, promote scenarios, or mutate snapshots.
In a mutable scenario, Astro switches to Scenario sandbox mode. That means it can draft any supported governed scenario-scoped change inside that scenario and keep every write scoped there.
Astro does not split writable scenarios into separate autonomy levels anymore. Once you are in a mutable scenario sandbox, Astro can plan broadly, draft bundles, and prepare larger governed scenario changes there.
Broad restructure asks such as reducing management layers while keeping monthly cost flat can now become phased scenario programs. Astro shows the current checkpoint first, keeps the rest of the phases in a compact timeline, and only drafts a governed bundle when the active phase contains real executable scenario changes.
Every execution step still waits for explicit confirmation, even in the writable scenario sandbox.

Permissions and limits

Astro access is separate from the data it can read.

You need the ai.assistant.read ability to see the assistant.
You need ai.scenarios.draft plus the same actor, placement, cost, team, department, org-unit, position, initiative, and scenario permissions used by the normal product flows before Astro can stage these scenario changes.
Astro still respects underlying data permissions.
If you can see a team but not cost data, Astro may answer the staffing part and say that cost data is unavailable.
For hypothetical move questions, Astro gives a read-only estimate from the current cited headcount records. Source and target stay as teams or departments, while people, agents, and robots act as actor-kind filters. It does not create a scenario delta or change anything.
Governed actor changes are blocked in Live baseline, snapshots, archived scenarios, and any other read-only context even if you can open Astro there.
Initiative proposals must stay linked to the current scenario, and scenario submission only moves a draft into human review.
The scenario sandbox never bypasses the hard rails. Astro still cannot promote scenarios, mutate snapshots, or write to Live baseline.
Tenant members, invitations, API tokens, and MCP connections remain out of Astro's write surface even though Astro can read them.
Department types, team types, sprint periods, role types, currencies, and AI pricing remain out of Astro's write surface even though Astro can read their current state or related plan context.
BambooHR connections, sync controls, and import execution or approval remain out of Astro's write surface even though Astro can read their current status.
Plan-gated follow-ups such as scenario history tasks, settings token management, MCP setup, integrations, and activity are omitted when the current tenant plan does not expose those surfaces.
Astro reuses the existing actor, placement, compensation, cost-event, team, department, org-unit, position, initiative, and scenario permissions when you confirm a proposal.

Astro does not:

make any direct change to Live baseline data
stage organisational model changes outside a mutable scenario
manage tenant members, invitations, ownership transfer, API tokens, or MCP connections through the write surface
run, pause, resume, disconnect, or approve BambooHR and import workflows through the write surface
approve or promote scenarios
restore snapshots
search external websites or external systems
use internal-only operator guidance as product documentation when answering knowledge questions

If a question is outside Orgonaut's supported domain, Astro says so directly.

Conversation history

Astro history is server-backed and tenant-wide.

Orgonaut keeps a unified persisted chat history per tenant and user. Conversations are not split by baseline, scenario, or snapshot — a single thread persists as you navigate across all contexts.
The Astro workspace sidebar lists previous chats and selecting one loads that conversation back into the main thread so you can review and continue it.
Astro groups workspace history into Today, Yesterday, Last 7 days, and Earlier so longer histories stay easier to scan.
The workspace sidebar shows the most recent unarchived chats first, caps the initial list to keep the sidebar manageable, and offers Show more when you want the full history.
First-turn chat titles are shortened into more intentional summaries such as roster, staffing, summary, and change-focused titles instead of repeating the full raw prompt.
Use the rename action in the workspace history rail when you want to keep an important thread under a clearer manual title; renamed titles persist across sessions, including the overlay's shared thread state.
Opening Astro on another device loads your persisted chats automatically.
Astro remembers the currently selected chat on the server per tenant and user, so the same thread resumes across devices.
On the dedicated workspace, the selected chat is also reflected in the URL as ?chat=<chat-id> so reloads, back/forward, and direct links reopen the same conversation.
Continuing an existing chat replays recent prior turns back into Astro so follow-up answers stay grounded in the same conversation.
The Astro workspace and the floating overlay share the currently selected chat.
Chat selection, thread loading, and archive behavior are server-owned, so conversations open consistently across browser sessions and devices.
Astro remembers pending confirmations, scenario batch follow-ups, and tray-dispatched visualization work so interrupted multi-step work can be resumed safely.
Moving between team, department, actor, and overview pages — including across baseline, scenarios, and snapshots — keeps the same selected chat.
Leaving the Astro workspace for another page hands the selected chat off to the overlay automatically.
Refreshing the page restores the selected persisted chat.
If a remembered chat is archived, Astro safely falls back instead of replaying the wrong thread.
When resumable work exists in another chat, Astro shows a Resume pending work banner with a direct link back to that thread and next step.
Use the archive button in the workspace sidebar to hide older chats from the active list without deleting them.
Chat age labels update live in the sidebar and use compact labels like now, 3m, 2h, or 5d.
New chat starts a blank draft immediately, but Astro does not create a persisted chat row until you send the first message.
Unsent drafts are ephemeral and do not become part of chat history until the first turn is submitted.
Orgonaut still records AI run provenance for each submitted turn.
Helpful and unhelpful reply ratings are stored on the relevant chat turn.

Good ways to use it

Start from the page you are already reviewing.
Use the dedicated Astro workspace when you want a longer session. Your conversation persists as you move between baseline, scenarios, and snapshots.
Ask one concrete question at a time.
Use the arrow-up composer button to send a question.
If Astro cannot reach the AI provider for a turn, it keeps you in the current chat, shows an inline retry message, and does not save a partial turn. Wait a moment and resend the question.
Use the clipboard action under a reply when you want to paste Astro's answer somewhere else.
Use the thumbs-up or thumbs-down actions under a reply to record whether that answer was helpful.
Use the cited source links to verify an answer before sharing it.
When the assistant asks for clarification, choose the exact team, department, actor, or scenario you meant.

Current v1 limits

Astro v1 is intentionally narrow.

It is not a general knowledge chatbot.
It does not replace scenario editing workflows.
It does not use provider-side memory or a global user profile memory outside the selected chat.
It may return a limited answer instead of guessing when sources are missing or the question is ambiguous.
Draft text does not become part of history until the first turn is submitted.

Search Results