Welcome to Taksim
The organisational intelligence control plane for coding agents. Govern access, delegate within policy, and verify outcomes with evidence.
How Taksim works
Taksim sits above the coding agents and intelligence resources your engineering organisation already uses. It does not replace the internal request-level planning of provider-native runtimes.
Capability, privacy, quality, and budget constraints determine which resources may be used before economics are considered.
Three distinct authorities
Native Auto
The provider or runtime owns internal selection.
Taksim Auto
Taksim owns the organisational resource decision where it has authority.
Delegated runtime
Taksim owns the policy envelope; the approved runtime owns its internal execution plan.
Local Connector, fail-open by design
Managed Claude Code and Codex sessions use a local Connector for policy decisions and evidence collection. Telemetry, persistence, and remote advice do not get to hold a healthy provider request hostage.
If the managed path cannot start safely, Taksim preserves the native client invocation instead of blocking the developer. Raw prompt, response, and source content is not persisted by default.
Verification is evidence-dependent
A verified outcome means the configured and available engineering evidence satisfied the applicable rules. It is not a guarantee that code is correct, secure, or complete.
Client compatibility
Last verified · 10 September 2026
| Client | History | Managed/live | Boundary |
|---|---|---|---|
| Claude Code | Available | Managed | Plain Claude remains independent |
| Codex | Available | Managed | Responses gateway; native model selection in v0 |
| Devin | Available | Managed | Native Devin remains the client |
| GitHub Copilot | Available | Not available | Historical visibility only |