Gloat is now Synq.Built for the AI era.Built for the AI era: HR that runs itself, so HR can do the people work.Read why
It routes every request to the right specialist, grounds the answer in your people data, acts on it inside your rules, and checks its own work — so HR gets answers it can trust and work that actually gets done.
Tell it to run Q4 reviews and it runs them: resolving who's in scope, reaching out in Slack, chasing the stragglers, and bringing calibrated results back — grounded in your records at every step, and stopping only where you set an approval.
Every answer it gives — and every action it takes on its own — runs through the same engine. Here's what each stage is built to do.
Comp, performance, policy, a messy employee case — each wants its own expertise. So a request doesn't hit one model: an orchestrator reads it and routes to the specialist agents that person is cleared to use, running one and returning the result or handing the whole conversation across to the expert. It decides again on every message, and scales its effort to how hard the ask is.
Before it replies, Synq reads your own records — searched by exact filter, full text, and meaning — and builds the answer from what's there. It has no path to the open web, so a headcount or a policy comes from your data, not the model's training. When the answer isn't in your data, it tells you instead of filling the gap.
People data is the most sensitive thing HR holds, so permission is engineered into every read — not checked after the fact. Each lookup runs as the person asking, on their own credentials, and their rows and fields are resolved before a single record loads; ask for one more field and you may get fewer rows back. A denial is indistinguishable from data that doesn't exist, so the agent can't be used to go fishing — and the same holds when you reach it through Claude.
Beneath your records sits a graph of people, their skills, your roles and work, and how much each skill matters to each. Synq reasons over it to answer the questions titles can't — who's ready for a role, where the gaps are, who to pair with whom — and ranks real matches of people to roles, projects, and mentors, rather than guessing from job titles.
Most HR work is a process, not a question — so Synq runs it. When a cycle or an onboarding opens, it stands up an agent for that run, messages each person directly in Slack, and drives the steps forward on a durable schedule, retiring itself when the phase closes. It reaches out unprompted — and nothing it can't take back happens without your approval.
An agent that can act has to be one you can bound. Every write Synq makes runs your rules first — allowed, held for approval, or blocked — and when approval is needed it parks the change and routes signed sign-offs to the right people, in order, with deadlines. It stays inside HR and inside your company, and does nothing irreversible on its own.
Every request executes on the asking person's own credentials — never a system key.
The access layer decides which records they may see before the query runs.
Sensitive fields are hidden in the same pass — down to a single value.
Tools reach your HR data alone. No open web, no other company's tenant.
Each change is checked against your rules, and held for approval when needed.
Actions are logged, and anything it can't take back waits for a person.
Your data never trains a model — ours or anyone's.
It's used only to answer your own team's questions, inside your own tenant.
No. The model is one swappable layer. The engineering is what's around it: an orchestrator that routes to specialist agents, retrieval that binds answers to your records, an access layer that enforces permissions in code, a skills graph that powers matching, and a workflow engine that runs processes under approval. Change the model and the architecture doesn't.
Every lookup runs as you — carrying your own credentials, not a service key. Row filters and field masks are resolved before the query runs, so the engine can't return a row or field you aren't entitled to. A denial is indistinguishable from data that isn't there, and if the access service is down the query returns nothing rather than more. The same enforcement holds when you reach Synq through Claude over MCP.
On defined workflow phases, yes. When a phase opens, the engine stands up an agent for that run, messages the assigned people in Slack, Teams, or Copilot, and advances the process on a durable, leader-elected scheduler — disposing of the agent when the phase closes. Proactive outreach is a per-agent setting, every write is gated, and nothing irreversible happens without a person's approval.
It reasons over a skills graph — people, their skills, your roles and work, and how much each skill matters to each — and ranks matches of people to roles, projects, and mentors. It's graph and vector matching over your own data, surfacing internal candidates and skill gaps, not a guess inferred from job titles.
Per answer, it returns whatever fits: prose when that's clearest, or working UI — a table you can sort, a record you can edit, a chart, or an action to approve — rendered inline, with the richer views on the web and a lighter set in Slack and Teams.
It isn't tied to one. Providers sit behind a single gateway and are chosen per task, so the engine can move to a stronger or faster model without a rebuild.