Where the work happens
Terminal, channel, and MCP entry points use the same AgenticLoop. The runtime owns tools, sessions, and execution controls.
A self-hosting agent runtime
Run the task. Inspect the result.
GEODE connects models to your tools, runs the task, and keeps execution records you can inspect.
Recorded public run
Create a TLS certificate
openssl-selfsigned-cert
GPT-6 Astra / highOne task, one trial
- terminal_exec20:29:32 UTCok
- terminal_exec20:30:27 UTCok
2 calls matched to 2 results. Command and result bodies are withheld.
Public event metadata. 9 payload bodies withheld; not a full replay.
The result becomes the next step.
The loop runs a tool, reads its result, then chooses another action or a final response.
- 1Your taskGoal and environment
- 2Choose an actionModel decision
- 3Run a toolPermitted action
- 4Read the resultContext for the next decision
Repeat when needed. A final response is not an external proof of success.
Approval policies, execution budgets, and cancellation govern the run.
Explore the runtimeHarbor × Terminal-Bench 2.1
Same model. Different harness.
GEODE and native Codex, measured on the same tasks and verifiers. Read the results with their limits.
Both runtimes used gpt-5.6-sol, max reasoning, and the OpenAI subscription route on Terminal-Bench 2.1. Harbor 0.22.0 supplied the same tasks and canonical verifiers. The comparator was native Codex CLI 0.145.0.
The GEODE arm used b549f3e's thin AgenticLoop adapter with a custom system_prompt_override and terminal_exec-only tools. This does not measure today's native/full-runtime configuration.
| Runtime | Passed / valid | Pass rate |
|---|---|---|
| GEODE | 339/429 | 79.02% |
| native Codex | 331/429 | 77.16% |
The task-balanced difference was +1.26 percentage points; the 95% task-bootstrap interval was -5.40 to +8.05 percentage points.
The preregistered full-suite primary is not measurable; the decision is inconclusive. The interval includes zero and does not establish superiority. This is a local paired-runtime diagnostic, not an official leaderboard score.
Exclusions and comparison limits
The protocol registered 89 tasks with 5 repetitions, or 445 trials per runtime. bn-fit-modify and tune-mjcf were excluded symmetrically because their amd64 verifiers could not run on the arm64 host. 6 native Codex trials remained infrastructure-invalid after their supplement caps.
Infrastructure-invalid trials were not converted into semantic zeroes. Harbor exposed no shared seed control. The run spanned several days; timing, provider capacity, and credential-principal effects cannot be separated from harness effects.
Run it. Measure it. Improve it.
Execution, evaluation, and scaffold search have separate responsibilities. A completed turn is not a promoted improvement.
Evidence for a judgement
Task results, trajectories, and verifier receipts support comparison and audit without becoming the same record.
Read the evaluation methodImprovement, tested separately
Scaffold candidates are searched and evaluated. An experiment's selected candidate is not a production release.
Explore the outer loopStart in your environment.
Install the stable release, connect a model, and give GEODE its first task.
Recommended. Install the latest stable release in an isolated tool environment.
uv tool install geode-agentAfter installation, run geode. Inside the session, use /login to review credentials. Choose an authentication path
Subscription authentication and metered APIs are separate routes. Model access and billing depend on your account and provider.