Get updates

// agentic data platform

The data platform where AI agents do the work.

WiseForge fuses a Delta Lake data platform with an agentic task board. Hand a task to an AI agent, watch it plan, edit and test live - then review and approve its work. Humans govern. Agents deliver.

Occasional build updates. No spam, unsubscribe anytime.

An agent run, replayed: a task is handed off, the agent plans against the linked data asset, edits and tests in an isolated sandbox, verifies the declared deliverable exists, then waits for human review and approval, with its cost and token usage on the record.

// how it works

Two worlds, one system.

Data tools don’t know about your tickets. Trackers don’t know about your data. WiseForge keeps the task, the tables it touches and the jobs it triggers linked - in both directions.

Data platform

A governed lakehouse: catalog, notebooks, jobs and pipelines over Delta Lake tables - Postgres for the hot path, ClickHouse for the analytics.

  • Browse and query a governed catalog - one RBAC, every read audited
  • Run notebooks and scheduled jobs against the same tables
  • Freshness, row counts and lineage tracked per table

Agentic board

A task board where work is picked up by humans or AI agents. Every card knows its data: the linked asset, the expected deliverable, the runs so far.

  • Hand any task to Claude Code or Codex - or keep it human
  • Watch the run live over SSE: plan, every edit, every test
  • Budgets, sandboxed execution and a deliverable gate on every run

// the defining moment

The agent handoff.

Delegation with a spine: a run you can watch, a deliverable that gets verified, and a merge that stays yours.

01

Hand it off

Pick a task, pick an agent, one click. Assignment stays cheap and reversible - the run is the deliberate, gated step.

02

Watch it work

The run streams live: the plan, every edit, every test, in a terminal you can actually read - not a spinner you have to trust.

03

Gate the result

Declare the table a task must produce and the run cannot reach review until it exists in the catalog, with rows.

04

Approve the work

The agent's run waits for your review. You approve it, or send it back with changes. Nothing lands without you.

  • Same rules as you. An agent can never do what the acting human couldn’t - one RBAC for both.
  • Data is never a command. Rows, logs and comments an agent reads can’t trigger actions - injection has no button to press.
  • Sandboxed by default. Runs execute in an isolated sandbox against a dedicated repo - never your production code.
  • Humans hold the merge. Every run ends at a review gate, and budgets cap what an agent can spend getting there.

// why we built it

Give your data team its time back.

Data teams drown in small, well-specified work: backfills, schema fixes, pipeline tweaks, one-off ingests - and a steady queue of data pulls for stakeholders. It is exactly the work modern LLM agents are getting genuinely good at.

But handing an agent your warehouse is a trust problem before it is a capability problem. An agent needs governance, a live audit trail, a budget - and it needs to actually see the data platform it is changing.

Most tools bolt AI onto one side: a copilot in the warehouse, or a bot in the tracker. WiseForge is built as one system, so the task, the data it touches and the jobs it triggers share one context - and one review gate.