On the roadmap: connect Claude, ChatGPT or your own tools to the ledger over an MCP server and developer API. Every action they take will run through the same autonomy thresholds, review queue and audit trail as the built-in agent.
The plan: point any MCP-compatible assistant at your workspace and let it read the ledger and run agent skills, governed by your firm's rules.
Planned: connect the assistants your team already uses.
Planned: build custom workflows on a typed SDK.
list, code, prepare, generate, all scoped by permission.
The design: tools called over the API or MCP will obey your confidence thresholds, surface low-confidence work to the review queue, and write to the audit trail. Filing stays human-gated.
API calls respect each skill's setting.
Every external action is recorded.
No tool can file without a human.
Connect an assistant or build on the API, under rules you set.