-
Notifications
You must be signed in to change notification settings - Fork 1
Configuration and Tuning
slipstream works with zero configuration. Everything below has a sensible default; you only touch it to change behaviour deliberately.
.claude/slipstream/dashboard.json in your project (src/dashboard/settings.ts):
Environment overrides, for a single session without editing the file:
| Variable | Effect |
|---|---|
SLIPSTREAM_DASHBOARD=0 |
disable the dashboard |
SLIPSTREAM_DASHBOARD=1 |
force enable |
SLIPSTREAM_DASHBOARD_OPEN=0 |
keep the browser shut |
SLIPSTREAM_DASHBOARD_OPEN=1 |
force open |
src/context/budget.ts holds the tunable constants:
| Constant | Default | Meaning |
|---|---|---|
BYTES_PER_TOKEN |
3.6 | bytes-per-token estimate; lower is more cautious |
DEFAULT_WINDOW_TOKENS |
200,000 | model context window |
COMFORT_FRACTION |
0.6 | the share treated as comfortably usable (ok below this) |
LARGE_FILE_BYTES |
16,000 | the threshold above which a whole-file read is flagged |
The window can be overridden per call: sp_budget(bytesRead, windowTokens) and slipstream budget --bytes N --window N. The levels are ok below COMFORT_FRACTION, warn up to 85%, compact above.
src/memory/recall.ts:
| Constant | Default | Meaning |
|---|---|---|
RECALL_TOKEN_BUDGET |
1,200 | the ceiling on the relevant subset reloaded at session start |
WEIGHTS |
branch 4, tag 3, file 3, description 2, haystack 0.5 | the ranking weights |
Raise the budget to reload more at session start, at the cost of more tokens; lower it to reload less. The weights are tuned against this project's own store (see Memory recall).
src/memory/digest.ts bounds the digest so it stays a memory the next session can afford to reload: MAX_DECISIONS 8, MAX_FILES 20, MAX_LINE 200 characters.
The statusline reads the same budget constants. To change what it shows, edit formatStatusline in src/statusline/index.ts (it is pure and unit-tested, so a change is caught by the test). To enable it manually, see Statusline.
slipstream writes only under .claude/slipstream/ in your project:
.claude/slipstream/
map.md map.json # the project map (from /slipstream:map)
memory/ # one .md per fact + MEMORY.md index
dashboard/<session>.jsonl # append-only event log per session
dashboard/server.json # the running server's pid/port/url
dashboard.json # optional settings
Add .claude/slipstream/ to .gitignore to keep it local, or commit memory/ if you want the team to share durable facts.
- Data formats for the exact shapes of these files.
- Memory recall for the recall tuning in context.
- Security model for what is written and what is redacted.
SarmaLinux . sarmalinux.com . Repository
Start here
Install paths
v0.6.0 features
- Map watcher
- Token forecast
- Replay export
- Configurable redaction
- Drift detection
- Per-skill opt %
- CI mode
- Lessons
Headline features
- MCP tools
- Observation memory & search
- Cross-IDE support
- Lossless compaction
- Memory recall
- Live agent dashboard
- Statusline
- Output style
- Subagents
Token efficiency
Skills
Internals
- Architecture
- Memory system
- Hooks
- Mind map and status
- Configuration and tuning
- Data formats
- Performance and benchmarks
- Design decisions
- Security model
- Testing strategy
Reference
SarmaLinux . sarmalinux.com
{ "enabled": true, // auto-start the local server on SessionStart "autoOpen": true // open the browser on first start }