CODING AGENT DEMO

Compare a coding task with and without Headroom.

Choose a bundled task and run two agents. Compare their code, test results, and context usage.

OPENCODE + LITELLM + HEADROOM

Try a task

Start with the bundled payment feature, or choose a GitHub repository.

Repository investigation

LIVE AGENT

Powered by OpenCode

OpenCode is an open-source AI coding agent. It reads files, chooses tools, and works with your model. Connect a repository to ask questions or edit code. The bundled payment task also supports restricted test execution.

About OpenCode ↗

Choose a task to see its goal and expected result.

README.mdevidence/incident.jsonservices/payments/config.jsonreleases.json

Agent task

The agent uses the tools allowed for the selected task. All branches start from identical files.

Include CCR (experimental)

Default: lossless Headroom keeps all records. CCR adds a third agent using sampled context with on-demand retrieval. Extra model calls can increase cost and latency; check the final answer.

Advanced Options

Compression threshold1,200 tokens

Model context window (tokens)Maximum output tokens per call

Use limits supported by your model. Follow-ups keep the original model settings.

Default: 1,200 tokens. Small requests pass through unchanged.

Input price / 1M tokens ($)Output price / 1M tokens ($)

Illustrative rates. Input uses a consistent estimate; output uses reported usage. Cache discounts and hosting are excluded.

Loading agent setup. This may take a moment…

Sequential, randomized order · same model

CLIENT-TRANSPARENT COMPRESSIONNo compression plugin

OpenCode tools → LiteLLM policy → Headroom → Model

File reads return to OpenCode. Its next standard LLM request passes through our proxy.

Agent experiment READY

Start a run to follow both agents.

Compression within policy-on run—

Total input before → after

Estimated task cost difference—

Baseline − compressed, input + output

Task latency difference—

Compressed − baseline

Exact evidence checks—

Baseline / Headroom · default task only

ⓘ

Agent paths may differ. Compare complete tasks as well as compression per request. Automatic OpenCode compaction and pruning are disabled. Repository sessions remain available briefly for follow-ups; reports and sessions have bounded retention and disappear on restart. These task checks are not a general correctness benchmark.

Compare a coding task with and without Headroom.

Agent experiment READY

Actual proxy requests

Before Headroom

Sent through LiteLLM

Repository file