Agents that act on your live operational state.
SixDegree doesn't just feed context to your agents. It runs them. Give an agent a goal and a scope, then run it on demand or attach a trigger so it fires when something changes in the graph.
Agents can observe, or take action. Every run is scoped, budgeted, and fully audited.
Goal
Each morning, flag accounts heading into renewal at risk and post a summary to Slack.
Plan
- Find accounts with renewals in the next 60 days
- Cross-check usage, open tickets, and last contact
- Rank by ARR at risk and write the summary
Scope · read
Report sent to #renewals
3 accounts at risk · $480k ARR · owners tagged
Anatomy of an agent
A goal and a scope. Nothing it can't reach.
A saved agent is a goal, the integrations and data it's allowed to touch, and the guardrails it runs under. Run it ad hoc, or attach a trigger so it runs on its own.
Goal
A plain-language instruction. The agent declares a plan before it acts, then traces the real relationships in your graph from question to answer.
Scope
Choose which integrations, tools, and entity types an agent can reach. It works inside that boundary and nowhere else.
Observe or act
Pick an autonomy level per agent. Observe reads and reports. Act writes back: update a record, open a ticket, trigger a workflow. Observe is the default.
Budget
Every run is bounded: max tool calls, tokens, cost, and wall-clock time. No runaway loops, no surprise bills.
Run-as identity
An agent executes under a real principal. Effective access is the principal’s permissions intersected with the agent’s scope. It can never elevate beyond the person it runs as.
Audit trail
Which agent, which version, who triggered it, what it touched, what it cost, what it concluded. Every run is durable and reviewable.
Three ways to run
On demand, on a schedule, or on an event.
The same agent definition runs however you need it. Click run, set a cron, or wire it to a webhook from any integration.
Ad hoc
Open an agent and run it now. Use it for one-off investigations, audits, or a quick second opinion grounded in live data.
Scheduled
Attach a cron trigger. A morning pipeline review, a nightly cost sweep, a weekly stale-account report. It runs while you sleep.
Event-driven
Fire on a molecule event, with optional conditions. New repo created, incident opened, deal stage changed. The agent reacts the moment it happens.
Safe by construction
Autonomy you can actually hand to an agent.
Governance isn't a setting you remember to turn on. It's how agents are built. Permissions are enforced at the tool boundary, not asked for in a prompt, so an agent cannot talk its way past your access rules.
- Observe, then act
Agents support both modes. Observe reads and reports; act writes back through your integrations. Most teams start read-only and turn on actions deliberately, gated by approvals for higher-risk writes.
- Scoped, not trusted
Effective access is the run-as principal’s permissions intersected with the agent’s scope. An agent can only ever do less than the person it runs as.
- Bounded blast radius
Per-run ceilings on tool calls, tokens, cost, and time. An agent stops at the budget, every time.
- Everything is auditable
Each run binds to a specific agent version and records the full trace. You can always answer why an agent did what it did.
Pairs with AI Persona
Every agent run speaks in your company's voice.
Attach lenses to an agent and your tone, ICP, policy, and priorities are injected into every run. The agent investigates with your judgment, not a generic model's.
Design Partner Program · 4 spots left
Put your graph to work.
We build the first agents around your stack and ship weekly. Direct founder access. Locked pricing.