From employees can use it to the enterprise can run it.
The gap is not the model. It is an organisational mechanism that lasts: capability stays in individual chats; knowledge enters the model without ongoing context; existing agents stay isolated; management cannot say who used what. AgentSeq connects entry, task, capability and governance into one system that keeps running.
Four layers
- EntryWeb workspace, IM channels, open API.
- TaskInstant and scheduled tasks, status, delivered results; a persistent task pipeline for long-running and recoverable work.
- CapabilityModels, skills, knowledge bases, memory, connectors, sub-agents — discovered and called within authorisation.
- GovernanceOrganisation, permissions, quotas, security, audit, operations data.
Four principles
- Autonomy and orchestration together. Generalist and specialist agents each take the role that fits, in the same task.
- Protocols first, multiple runtimes. Skills and A2A extend the native runtime; other agent runtimes are called through ACP.
- Brain and hands apart. Agents reason and coordinate; a sandbox executes code and tools. Each scales independently.
- State outside the process. Sessions, memory, task state and workspaces survive any single run.
How a task runs
Set the goal → identify the boundary (organisation, member, authorisation) → assemble context (knowledge, memory, business data) → select capabilities → execute collaboratively → consolidate → deliver to Web, IM or the business system → record status, usage and audit.
one entry, clear status, a reusable result.
assembly, selection, collaboration, consolidation.
who, what, how much, and how to trace it.
Knowledge and memory
Knowledge answers what the enterprise knows; memory answers what this task continues. Both enter the task context under organisational boundaries: knowledge scoped by organisation, member and business area; memory — personal, team, organisational, per business object — written and read according to the task's subject and authorisation. Knowledge can be supplied by DataSeq.
Bring your existing agents
| If the capability is… | Path | It becomes… |
|---|---|---|
| clear-edged, standardisable | Skill | an enterprise skill in the capability catalogue |
| specialist, able to run autonomously | A2A | a managed sub-agent that collaborates in complex tasks |
| already a complete service | ACP | a platform agent service that authorised users call directly |
Operations and governance
A team console puts organisation, capabilities, resources, activity and risk on one screen — different roles see different questions (business owner, administrator, IT, security audit). Governance is a loop: register assets → authorise → execute inside the boundary → record → observe → audit → adjust policy — and back to the start.
Team overview
Activity by team tasks per day · last 7 days
Most-used capabilities calls · 30d
Spend by team USD · 30d
Recent audit events who · what · under which authorisation
| Time | Actor | Action | Object | Scope | Result |
|---|---|---|---|---|---|
| 14:02 | j.chen | task.run | Research brief | Research / analyst | completed |
| 13:47 | admin.wu | policy.update | External sharing | org / admin | applied |
| 13:31 | m.li | agent.call | Contract reviewer | Legal / counsel | completed |
| 13:12 | s.park | knowledge.read | Pricing policy v4 | Sales / rep | denied · out of scope |
| 12:58 | scheduler | task.run | Weekly digest | Operations / service | completed |
Solution blueprints
Enterprise research & knowledge service
Research, Q&A and content within authorised knowledge.
Specialist agents in concert
Existing agents keep their way of working and join complex tasks as managed sub-agents.
Agentic capability inside business systems
Portals and workflows create tasks over the open API without building their own agent platform.
Getting started
Asset inventory → scenario selection → solution design → integration → pilot → continuous operation.
Deployment form, model access, data boundary and operational responsibility are designed together, not taken from a template.