Product
An agent you talk to. It does the work.
HAXAL is small on purpose. No sprawl, no forms to fill. You describe what you want; the agent declares it, runs it, and watches it. You chat to get answers, to change what it tracks, or to have it execute and schedule work. Configuration is a conversation.
The lifecycle
Connect
Point the agent at your sources. Each is added by secret reference, by name. The agent reads them; the key stays out of the spec.
Describe
Say what you want. The agent writes and maintains the spec. The board renders from it, and so do the schedules and the alerts.
Operate
It runs, schedules, watches, and escalates. Every run is governed by the Trust Harness and recorded forever.
Agents, not screens
Chat is the primary surface. The board is the artifact.
Each agent owns a view, the sources it is allowed to use, its threads, its schedules, and its own memory and audit log. You interact through conversation, and the agent maintains the spec for you.
Separate, logged chats
Each agent has many threads, like separate conversations. Each is an append-only, ordered log of messages and runs. Auditable forever.
It accepts work from three sources
From you, from other agents, and from triggers: a schedule, a webhook, a threshold crossing. It accepts, queues for approval, or declines by policy.
The spec is editable
The agent maintains the spec from conversation. Power users can edit it directly. No-code and low-code in a single path.
Trust Harness
Every run is inspectable end to end.
Inputs, tool calls, the model exchange, and the verdict are all on the record. Generation and validation are kept separate, on purpose, so the check is independent of the answer it is checking.
Every answer traces back to its source
No claim ships without a source. No fabrication, by design and by policy.
A number on every reply
Each run scores its own confidence. Below the gate, it routes to a human.
Traceable to the row
Follow any answer back to the exact source, query, and value it came from.
Autonomous execution
Work fires while you sleep.
Per-agent schedules, long-running agent loops, and a runner model for machine-bound work. Short jobs run light. Long model loops run supervised. Results post back into a thread, on the cadence you set.
"schedules": [ { "every": "10m", "action": "refresh-data" }, { "every": "week", "action": "post-summary", "target": "thread:weekly" }, { "every": "hour", "action": "evaluate-attention", "target": "thread:alerts" } ]
Always-on, on your cadence
Set the rhythm in plain language. The agent keeps the work running without you in the loop.
Long loops, bounded
Every loop is bounded and checkpointed. Nothing runs away. Nothing stalls silently.
Results come to you
Output posts into the right thread. The alert arrives with the answer already attached.
Governance and multi-tenancy
From agent sprawl to a control plane.
As a team spins up many agents, HAXAL is the layer that governs them all: isolation, roles, audit, kill-switch, and data residency. Identity says who you are. The control plane says which tenant and what role, and trims every query to it.
Per-tenant isolation
No shared rows. Strong database and instance isolation from day one.
Roles and query trimming
Roles and per-tenant scope are enforced at query time, on top of your verified identity.
Audit and kill-switch
Every run logged. Stop any agent instantly. Human-in-the-loop is the default posture.
Describe it. The agent declares it.
Tell us the work your team repeats by hand. We'll stand up an agent that runs it.