Skills and assistance
Planned
What it is
A person can declare a skill, record a certificate, and see a suggestion that comes from work they have already finished, together with the evidence behind it. When work needs an owner, people are filtered by skill, availability and permissions before anyone is ranked, and a manager approves the assignment. A brief can become a draft breakdown, and a weekly summary is not sent until a manager approves it. The product does not infer health, personality or intent.
AI features at a glance
| Feature | What it does | Where you find it | More |
|---|---|---|---|
| AI assistant | Answers questions about your records and prepares changes you confirm | Ask AI in the top bar, ⌘J / Ctrl+J, or the Assistant page | AI assistant |
| Search in plain language | Turns a question into ordinary search filters | Search under My work, or ⌘K / Ctrl K | Searching work |
| Planning and staffing | Answers capacity questions, suggests people, drafts plans and breakdowns | The assistant (/plan, /staff, /free) and Break down on a work item | Planning and staffing |
| Skill signals | Suggests raising a skill by one level from finished work | Your skills page; admins use Settings › AI › Skills | Skill signals |
| Writing help | Writes, shortens, expands or changes the tone of your text | ✦ Writing help in description, comment and reply boxes | AI assistant |
| Thread summaries | Summarises a ticket and its replies | ✦ Summarise thread on a ticket | AI assistant |
| Night notes | Prepares a project summary, schedule flags and a timesheet note | The project page and Insights | See below on this page |
| Access and tokens | Decides who can use AI and tracks AI token use | Settings › AI › Access & usage | AI access and tokens |
Where the models run
- On your own servers. In an on-premises installation, every model runs inside the institution and no data leaves it.
- In Akollo's cloud. Quick checks and search run on small models we host. Longer answers, such as a project plan, may use a large model from a cloud provider. Those requests go only to providers that keep no copy of your data and do not train on it, and only to the providers your administrator allows.
- The right size of model for each task. Quick checks use small models. A project plan, a work breakdown or an explanation of a staffing suggestion uses a larger one.
- Restricted data never leaves. Data classified as restricted is answered only by a model inside the installation. If there is none, the request is refused rather than sent elsewhere.
- Your administrator decides the highest classification a cloud model may receive.
- Every AI request is recorded with the model it used and how much it used, never with the question or the answer.
Note
The record of each AI request holds the model and the amount used. It never holds the question or the answer.
Selection checks
Once a month, the product compares how often people on different teams were chosen for a staffing recommendation. It can also compare seniority bands when those bands are recorded. It does not group anyone by health, gender or a similar trait, and it does not keep names in the check. When the comparison is uneven, administrators are asked to look at assignments. The section on the usage page arrives with that page.
Night notes on a project
During the night, the product can prepare three suggestions for the next morning:
- a short project summary,
- a closed list of schedule flags,
- a note about one timesheet day whose hours do not match what was expected.
A manager reads them on the project and on Insights. They are suggestions. Every number in them comes from that project's own records. Nothing is sent onward, and a person still approves any change to the work.
Search within records you can open
A search suggestion names only records you can already open. It does not add a record from another organisation, and it does not open a record you cannot open. When two records share an identifier, the suggestion names only the one you can open. When a record is removed, it drops out of search suggestions. There is no separate search page. The suggestion names the records. It does not replace them.
FAQ
No. Assignments are approved by a manager, a weekly summary is not sent until a manager approves it, and night notes are suggestions only. A person still approves any change to the work.
No. Data classified as restricted is answered only by a model inside the installation. If there is none, the request is refused.
No. Each AI request is recorded with the model it used and how much it used, never with the question or the answer.
No. The product does not infer health, personality or intent, and selection checks never group people by health, gender or a similar trait.