Context
Repository-aware sessions
Project context, files and durable conversation state stay attached to the engineering task instead of living in an isolated chat box.
Flagship product · Trebell Code
A coding harness
for real engineering work.
Trebell Code gives AI coding agents a durable desktop workspace around the model: repository context, terminals, files, Git/worktrees, browser automation, recovery, verification and telemetry.
Open sourceApache-2.0Windows · macOS · LinuxProvider-independent
What it provides
The model is only one part of the system.
Trebell Code focuses on the engineering environment that turns inference into repository work developers can inspect, verify and recover.
Execution
Terminal, files and Git
Agent workflows can edit files, run commands, work with Git/worktrees and interact with development tooling.
Verification
Evidence before “done”
Tests, builds, browser runs, screenshots and explicit verifier outcomes are treated as evidence, not optional decoration.
Runtime
Trebell Native + external harnesses
Trebell has its own native route while retaining compatibility with external coding runtimes for comparison and workflow choice.
Models
Provider-independent
The model provider and the agent runtime are separate decisions so users can change one without rebuilding the entire workflow.
Recovery
Designed for long-running work
Durable state and recovery reduce the cost of interrupted sessions and make longer software tasks easier to resume.
Proof
Measure the finished result and the actual cost.
Trebell benchmarks the same model across different coding harnesses and compares verifier outcomes alongside input/output usage and dollar cost. The goal is not “fewest tokens at any cost”; it is equal or better software quality with less wasted inference.
Benchmark evidence lives on the product site.
Results are shown with task context, verifier outcome, model/reasoning settings and cost measurements so the comparison can be inspected rather than treated as a magic score.