A desktop workbench for ServiceNow developers. Sign in the way you always do, then write scripts, trace records, review update sets and test flows across all your instances. Nothing installs on the instance.
the workbenchscript include · record form · findings clean · 7 references
35
artifact types in one editor
1.4 s
where_used over 12,901 artifacts, zero live requests
8
deterministic security checks, each with evidence
~5,100
automated tests, plus end-to-end runs on a live instance
script editor
An editor that knows the instance.
Completion comes from the instance’s live dictionary. The validator flags unknown tables, invented Glide classes and GlideAjax mistakes on open and on save, and offers a quick fix where one is safe.
7 references · before-insert on 4.1M rows · 3 ACLsblast radius, computed on open and save, no model involved
business rules, script includes, ACLs, catalog items and more
F12 definition, Shift+F12 usages from the local code graph
version history with a real diff and gated revert
compare one artifact across instances by name
dependenciescalls · property · roles · tables · 1 hop
lists and forms
Lists and forms, as the instance has them.
Lists open in the views configured on the instance, with columns sized to what they hold and references you can click through. Forms follow the form layout of the view you pick.
the instance’s own views in a view picker
encoded query and quick filters above every list
choice labels in full, dates in monospace
sections and columns read from the form configuration
a listdefault view · references open · 77 showna formform layout from the instance · default view
records and the why
Ask a record why.
Open any record and see what runs on save, in order. See why a field is mandatory, which ACLs decide access, and which business rule most likely changed a value, ranked from the audit trail. All fields shows the rest: raw values, where each field is declared, and what the form leaves out.
Cmd+K takes a sys_id, INC0010001, a URL or an encoded query
/br /acl /logs /dict and 14 more slash commands
act as another user in a separate side session
background scripts that clean up after themselves
record explorerwhy mandatory · business rules on save · in orderall fieldsraw value · display value · declared on
update sets
Update sets, reviewed like pull requests.
Every entry gets a diff, every payload script goes through the validator, and dependencies are checked against the target instance before you promote. After an upgrade, skipped records are sorted for you.
collisions with other open update sets
capture audit: ACL without roles, field without table, cross-scope
dependency check against the target before promotion
skipped records sorted: OOB-only, customer-only, conflict
update set reviewper-entry diff · validator finding · fields changed
Every write goes through one gate. You confirm it, vikstra reads it back, and the old version is kept.
UI, script runner, AI agent, MCP: there is no second path.
confirm
snapshot
write
read back
audit
flows and ATF
Watch a flow run.
Flows are drawn as a graph, with branches side by side. Run one and the path lights up as it executes. Compare two runs step by step to find the branch only one of them took.
add and drag steps, data pills, publish with a backup
replay a run, diff two runs, follow recent runs
generated ATF tests, kept only if they pass and leave nothing behind (needs a model)
flowson-call: time off approval · if · else-if · steps
analysis
Search every script, locally.
vikstra keeps an encrypted index of each instance on your machine, with full-text search over every script body. A where-used search runs against that index and sends no requests to the instance.
sweeps ranked against the out-of-box baseline
security review: eight checks, each with evidence
ACL matrix: users by operation, never probed by writing
dependency graph with rename impact
analysiselevated roles · public endpoints · script findings · confidence
AI, if you want it
A model that is checked against the instance.
The editor, records, update sets and analysis work without a model. Add one and it reads the live instance through about 70 read tools. A sys_id in an answer that no tool returned is flagged, and scripts it writes are validated against the live dictionary. A personal-data filter runs on every request and cannot be switched off.
Anthropic, OpenRouter, OpenAI-compatible, local Ollama
scripts checked against the instance before “verified”
plan and apply: approval bound to the plan hash
MCP server and client, same gate, same audit
user story mode, on a local model, writes nothing
chatchecks passed · tool calls · open in script runner
safety rails
Careful by default.
production
Read-only until you allow writes.
Write access lasts one session. Dry runs and test generation are refused on prod.
every write
Confirmed, then read back.
A push is refused if the record changed since you opened it.
history
Snapshotted and audited.
The prior state goes to an undo store, the write to an append-only log.
at rest
Encrypted, on your machine.
Chats, tabs and snapshots in ~/.vikstra are AES-256-GCM encrypted. The index is sealed on quit and screen lock.
Nothing. vikstra uses the session from your normal sign-in. When background scripts are not available to you, you can run a script as a temporary fix script instead, which vikstra then deletes together with its update set entry. Anything it could not remove is reported.
What does it need from my instance admin?+
Nothing. You sign in through the usual login page, SSO and MFA included. There is no API account, no OAuth app and no integration user. vikstra can do what your own account can do, and nothing more.
Where does my data go?+
To your disk, in ~/.vikstra. Chats, open tabs, jobs and snapshots are encrypted with AES-256-GCM, key in the OS keychain, and the index is sealed the same way when the app quits or the screen locks. The audit log, flow backups and, in the default workspace mode, a copy of the instance’s scripts are plain files. There is no vikstra cloud and no account, and no telemetry, crash reports or analytics. Besides the instance and the AI provider you configure, the app contacts an update feed if one is set, Hugging Face for the privacy model when no filter service is set up, and GitHub when you download the ServiceNow docs.
Do I need AI to use it?+
No. The editor, validator, record explorer, update set review, flows and analysis all work with no model. Plain-language list filters and ATF test generation need one. Each instance can be set to hosted AI, local only, or no AI.
Can it delete records?+
Only the child rows of a composite artifact, such as a catalog item’s variables, through the same gate and with a snapshot. The AI agent, MCP clients and the scripting host cannot delete at all.
What can’t it do yet?+
UI Builder and Workspaces, decision tables, custom Flow Designer actions, an import set test bench, and Service Portal widget preview (widgets are edited as scripts). For those, the platform UI is still the right tool.
Which platforms, and can I download it?+
macOS. A Linux build exists but has seen little testing, and Windows is not supported yet. It is version 0.1 and builds are not public yet. The macOS build is not notarized, so the first launch needs right-click, Open. Write to us and we will send you one.
early access · v0.1
Try it on your own instances.
vikstra is early. Tell us what you work on and which instances you look after, and we will send you a build.
