Amidship-owned production platform

Agentify keeps state and control in the application while AI agents do the work.

Agentify is an application-owned operating layer for agentic workflows. It coordinates models, tools, knowledge, background work, human checkpoints and workflow state without making the model the system of record.

Different workflows can use different capabilities while sharing the same governed operating foundation.

Amidship-owned production platform

The problem

Different AI workflows need different controls.

A public AI front desk needs grounded answers, limited actions, structured intake and a clear human handoff. A coding agent needs far broader execution access, but it also needs isolated workspaces, durable run state, pull-request and CI/review tracking, retries and a final human merge decision.

Build each of those as a one-off loop around a model and the difficult parts get rebuilt every time: state, permissions, knowledge, failure handling and human control. Agentify keeps those responsibilities in the application and gives each workflow only the capabilities it needs.

Key decision

Keep state and control in the application. Give each workflow only the capabilities it needs.

Application state

  • Task + messages

Replaceable capabilities

  • Model gateway
  • Authorized tools

Control paths

  • Human intervention
  • Pending · failed · completed
Agentify keeps task state and control paths in the application while models and tools remain bounded, replaceable capabilities.

Architecture

The system around the decision.

The shared Rails foundation persists tasks, messages and tool calls; routes model selection through account configuration; assembles authorized tools; handles background work, context limits and failures; and provides a substantial knowledge system with chunking, contextual embeddings, HNSW vector indexing, hybrid retrieval, filters, rollups and import/export operations.

The Front Desk uses that foundation with a restricted public configuration: selected knowledge bases, deployment-specific intake and handoff tools, persisted leads and transcripts, rate limits, and no arbitrary agent tools or memory. The coding-agent workflow uses the same application principles with broader execution access inside BranchBox-isolated workspaces while Agentify tracks the run through pull request, CI, review, and human merge.

  1. 01Application-owned task + message state
  2. 02Account-level model routing
  3. 03Authorized tools + human control
  4. 04Knowledge / RAG system
  5. 05Background work + failure recovery

One shared foundation, configured for different workflows.

Restricted public workflow

A configurable anonymous AI front desk can use only the knowledge and actions selected for that deployment, capture structured intake, persist the conversation and escalate to human review.

Broad execution, isolated workspace

Claude Code or Codex can work inside an isolated BranchBox workspace while Agentify owns run state, pull requests, CI/review loops, retries, session continuation and the final human merge decision.

Authentic product evidence

The operating boundary, visible in the product.

  1. Orchestrate
    Agentify Orchestrator showing multiple persisted coding-agent runs in distinct lifecycle states.

    Persisted coding-agent runs remain visible across active, CI, review and ready-to-merge states.

  2. Keep the human boundary explicit
    Agentify run detail showing a ready-to-merge state and explicit final human decision boundary.

    A green pull request pauses at ready to merge, returning the final decision to a person.

  3. Configure the workflow
    Agentify Front Desk configuration showing published status, intake fields, knowledge, hours and human handoff controls.

    The Front Desk exposes its model, knowledge, intake, operating hours, next step and handoff destination as application configuration.

Deterministic Agentify captures · August 2026 Real Agentify routes rendered with seeded synthetic state. The images are exact copies of the reviewed product captures and contain no customer data.

Why this matters

The model is only one part of the product.

Agentify reflects how Amidship approaches operational AI when the workflow has to remain understandable and controllable after the first model call. The application keeps durable state, knowledge, permissions, actions, failures and human decisions explicit, so AI capability can change without making the entire product opaque.

  • Persist workflow state outside the model
  • Treat tools and knowledge as explicit, configurable product capabilities
  • Give public and internal AI workflows different permissions and controls
  • Use human review as part of the system design rather than an afterthought
  • Keep provider/model choice replaceable without giving the model ownership of the workflow
  • Build retrieval and indexing as operational infrastructure, not a prompt-time demo

Production platform

Inspect Agentify directly.

Open Agentify (opens in a new tab)