New chat
Mock cards
Workflow / Subagent (fixture mock)
Trace samples (§1.1 / §1.3)
Viz P0 gap fills (viz-coverage-matrix §5)
Clickability audit fills (docs/demo-clickability-audit.md)
Field P0 收尾批 (§3 P0 剩余)
Where should we start?
DSH is a workbench with the agent's internals in view. Any turn can be branched, any plugin can be written on the fly, and you can always see what the model still remembers.
Switch to Graph to see this session's turn DAG.
Switch to 时序 to see this session's step timeline.
Switch to Trace to see this session's step tree.
Switch to Log to replay this session's full event history.
Session Tree
Every session as a branchable timeline. Fork points, subagent lineage, and status — all in one view.
fork
subagent
No sessions yet. Start a chat, then use Fork on any turn.
Select a session
Click a row on the left to see its metadata, timeline, and actions here. Hover any row for quick actions like fork or compare.
—
·
idle
Timeline
turn · tool · compact · fork
No events yet.
turn
tool
compact
fork
Compare branches
—
—
—
—
Fork point highlighted; the two timelines diverge after that seq. Click any marker to jump to the source turn in that branch.
Missions
Long-running tasks across all sessions. Watch progress, jump into any subagent.
Mission demo
Plugins
Manage the capabilities available to the runtime. Toggle on to enable; add your own or install from the catalog.
| Enabled | ID | Package | Source | Runtime |
|---|
Playground — scratch runtime, unsaved
Experiment with overlay edits in isolation. Apply commits the changes; discard tears it down.
Compare against session:
Live history (read-only)
Playground (this run)
Pull Requests
Not loaded
Evals
demo · G1 pending
Rubric library, evolution over time, and per-rubric rollout grid — three views over one scoring event log.
Rubrics
demo · G1 pending
28 subtasks · 7 groups
Growth
Our companionship log is a set of auditable facts, not persona copy.
Activity heatmap
less
more
Runtimes
Local profiles, daemon state, adapter capabilities. One row per profile — expand for the capability matrix.
composed locally · gap G8
Settings
Demo-tier price overrides and API-key presence. Never reads or displays a key value.
Model pricing
Rates are in USD per million tokens. Edits persist to this
shell's
localStorage and flow through
window.__dshEffectivePriceTable(); the
runtime is never told. Press Reset to remove an
override.
| Model | Input | Output |
|---|
API keys
Presence only — this page never shows a key value. A
personal key is one you own directly (your
own DeepSeek / OpenAI / Anthropic API key); a
service credential is a shared token
managed by an operator. To change a key, edit the launch
environment or your profile YAML — the shell does not
write keys.
| Name | Tier | Description | Presence | Last used |
|---|
Optional pages
Enable extra pages in the left sidebar. Hidden by default
to reduce the first-run surface; toggles persist to
~/.dsh-desktop/config.json under
hiddenPages. Advanced users can hide any page
by adding its id to that array manually (e.g.
["prs","growth"]).
Bench
demo · G18/G19/G20 pending
Local-lite experiment runner — matrix / plugin A/B / repetition. DSBench-compatible
code_result.json.Context
Per-turn ledger of shadowing, injection, recall, and compact events.
Interventions
no interventions
Load a sample session to see the page shape, or start a chat from the Chat tab and come back.
Tracing
All sessions in this project — latency, tokens, cost, error rate.
No sessions yet.
Open a chat from the Chat tab, or seed a fixture session via DevTools (
window.__dshQaSeedSession when DSH_QA=1).