Open source

Engineering in public.

Engineering credibility should be demonstrated rather than claimed.

Argbit maintains open-source tooling in the areas where enterprise AI is hardest: orchestration, evaluation, integration and observability.

Repositories below are published as they reach a usable state.

Repositories

Tooling we use in client engagements.

Each repository exists because we needed it on real work. Status reflects the current state honestly rather than aspirationally.

argbit/agent-runtime

Orchestration runtime with typed tool contracts, per-run budgets, retries and structured traces.

TypeScriptIn development
argbit/eval-harness

CI-executed evaluation suites for agentic systems: task completion, escalation precision, latency and cost per task.

PythonIn development
argbit/mcp-connectors

Model Context Protocol connectors for enterprise systems of record, with identity propagation and audit records.

TypeScriptIn development
argbit/trace-viewer

A reader for agent traces designed for support engineers rather than for researchers.

TypeScriptPlanned
Why we publish

Public code is the clearest engineering signal.

Reading an evaluation harness or a tool gateway tells you more about how a team thinks than any capability deck.

Contribution principles

  • Small, documented interfaces
  • Evaluation before features
  • No vendor lock-in by default
  • Security and auditability treated as requirements
  • Honest status labels on every repository

Using one of our repositories?

Issues and design discussions are welcome. If you are evaluating the tooling for a production programme, we are happy to talk through the trade-offs.

Start a conversationNo AI theatre. Just engineering.