From console to pilot

1. UI

   Web console

   Done

   • ✓Task composer, task log, timeline
   • ✓Approval gate, budgets, diff viewer, artifacts
   • ✓Workspaces, models, tools registries
   • ✓Docs, guide, about, security, roadmap and FAQ
   • ✓Five interface languages, motion and page transitions

   deliverable · Full console UI on a mock runtime.

2. Sprint 1

   Foundation & inference

   In progress

   • ✓FastAPI control plane, task schema, state machine
   • ✓Model adapter (OpenAI-compatible) for vLLM / SGLang / Ollama
   • ✓Dev auth stub with workspace authorization
   • ✓Console API mode: live SSE timeline, approvals, cancel, resume
   • ○Docker Compose with Postgres
   • ○Latency / memory benchmark on a GPU host

   deliverable · Chat backend connected to the local model.

3. Sprint 2

   Agent runtime & repository tools

   In progress

   • ✓Planner with schema-validated plan and bounded repair
   • ✓Read-only repo.tree / code.search / file.read with traversal protection
   • ✓Persisted steps and tool results; resume from persisted state
   • ✓Supervised plan approval bound to the exact plan
   • ○Measure tool-call and plan validity on a real Qwen model

   deliverable · Agent can inspect a repository and produce a validated plan.

4. Sprint 3

   Coding execution

   Planned

   • ○Isolated workspace and patch application
   • ○Test / lint execution
   • ○Bounded retry and replanning

   deliverable · Agent can produce and validate a patch.

5. Sprint 4

   Security & recovery

   Planned

   • ○Policy engine and approval records
   • ○Sandbox hardening and audit logs
   • ○Isolation and prompt-injection tests

   deliverable · Controlled execution with observable safety behaviour.

6. Sprint 5

   Benchmark & pilot

   Planned

   • ○Repeatable coding evaluation set
   • ○Success-rate, latency and resource measurements
   • ○Limited developer pilot

   deliverable · MVP with documented evaluation results.

The 10-week MVP timeline is a planning estimate; actual duration depends on hardware, scope and security requirements.