Agentic applications

  • TypeScript / Python
  • MCP tool interfaces
  • Workflow state
01

FullInfill CRM

Application study ↗

Business software & tool interfaces · work at RAM

Records, drafts and board actions share an application with the assistant tools that operate on them.

CRM trigger and agent graph with scoped outputs
Actual application capture · event, schedule and tool-scope graph.
Full resolution ↗
  1. Event
  2. Scoped tools
  3. Record / draft
  4. Review
Behind the work

The React application connects customer work, campaigns, boards and mail. Its MCP interface publishes the application’s tool schemas and routes calls through the same tool executor, with request context and workspace scope.

02

Praxis

Implementation note ↗

Experimental capability resolver

Choose the model’s working context before the call, then leave a record of exactly which configuration was used.

PraxisLocal CLI capture
$ node bin/praxis.mjs resolve \
    --model openai/gpt-5.6-sol --intent build

  manifest hash  px_0cf8c0b980b90cfd
  model          openai/gpt-5.6-sol
  intent         build
  effort         high
  skill depth    OPEN
  tool budget    20
  capability     UNVERIFIED
Actual local resolver output · shortened; no model call.
Behind the work

The resolver combines model capabilities, intent, tool budget, effort and policy into an inspectable manifest. A stable hash identifies that configuration. The prototype keeps configuration and session records inspectable.