Docs
A quick tour of SeekSpeed Terminal v1.
1. Connect a backend
Go to /app/connectors. Add an OpenAI-compatible endpoint (or use one of the presets for OpenAI, Anthropic, OpenRouter, vLLM). Hit Test to validate. Credentials are stored in your local workspace.
2. Create an agent profile
An agent profile is a lightweight configuration layer: system prompt, model/connector, tool usage mode, memory mode, token budget. Profiles let you benchmark the same prompt across configurations.
3. Run a benchmark
In /app/benchmarks, define a prompt set (inline or paste CSV/JSON), pick a profile, and run. Results capture TTFB, total ms, tokens, cost estimate, error rate. The detail view shows per-prompt timings and the response.
4. Review recommendations
After each run, the rule-based optimization engine generates recommendations with expected impact, difficulty, confidence, and apply guidance.
5. Export a report
From /app/reports, save a snapshot of a run with notes. Export to JSON or CSV for sharing.
6. (Experimental) Spec Lab
For self-hosted / open-weight stacks only. Configure draft/target model pairs and dataset; the orchestration shell creates a job object and streams logs. A real DeepSpec worker can be plugged in via the adapter layer.