The OmniLore approach
AI work should feel capable — and still make sense.
The four principles below describe the promises we try to make concrete: better evidence, clearer permission, meaningful separation, and work that can pick up where you left off.
Grounded
OmniLore starts with the best available evidence for the task, then makes the boundary of that evidence visible.
What it means
Approved organization guidance and cited public sources come before confident-sounding invention. If the workspace cannot find support, the answer should say what is missing.
What you can expect
Ask a product question in Chat, search an approved knowledge corpus, or inspect a permitted repository. Sources and result receipts help you decide what to trust and what to verify.
Example: Instead of presenting a guess as policy, OmniLore points to the relevant guidance, summarizes it, and leaves consequential decisions with the responsible person.
Permissioned
An agent is useful because it can act, but the action has to belong to the person, role, workspace, and connection asking for it.
What it means
Tools and information are filtered by the signed-in member, role, workspace, and current feature policy. Consequential actions can require an explicit confirmation.
What you can expect
RepoHub, CodeHub, connectors, and agent actions stay inside the access granted to that member. A denied request should fail closed rather than quietly broaden access.
Example: A code agent can propose a change in an authorized repository, but merging or deploying remains a separate, reviewable action with its own boundary.
Separated
Private work is not one giant search index. OmniLore keeps identity, workspace, and connected-source boundaries meaningful.
What it means
Personal conversations, session summaries, connected sources, and workspaces are not ordinary cross-member search. Organization access is granted intentionally, not assumed.
What you can expect
Support, recovery, security, and lawful-access paths are limited administrative routes — not a general invitation to browse another member's work.
Important boundary: Local-first coding can keep ordinary work on operator-controlled hardware, but OmniLore is not automatically air-gapped. A connected request can send required content to OmniLore services or the external provider selected for that request.
Continuous
Good work should not disappear when a tab closes, a handoff happens, or the next step belongs to another surface.
What it means
Structured handoffs, intent, permitted personal memory, project context, and durable receipts help work continue with less repetition.
What you can expect
Start in Chat, continue in CodeHub or CoTrinity, and keep the code in RepoHub. The system can carry context forward without pretending that an informal memory replaces an approved record.
Example: A coding task can leave a clear handoff with the files touched, checks run, and unresolved questions so the next person or agent knows what happened.
How the four principles work together
- Grounded: find the evidence and name its limits.
- Permissioned: check whether this member and workspace may use it.
- Separated: keep unrelated people, sources, and workspaces out of the context.
- Continuous: leave the next person a usable handoff or receipt.
See the principles in the product
Chat is where questions begin. CodeHub and CoTrinity help with repository work. RepoHub keeps the code and review trail private for invited members.