Native REPL

Slash commands, tools, sessions, skills, and the training pipeline — everything you do inside owlcoda.

Type owlcoda and you land in the native REPL. It looks like a terminal because it is one.

Top slash commands

Command What it does
/help List every slash command, grouped
/model Open the model picker; switch the active model
/cost Token + cost breakdown for the current session
/budget Used vs estimated context window
/perf Per-model latency + tokens/sec + success rate
/doctor Recent errors, uptime, trace status
/config Active model, proxy, router, flags, session info
/trace Toggle JSON request/response tracing
/tokens Cumulative input / output / total token counts
/sessions List recent sessions; resume any
/tag <name> Tag the current session
/branch Fork the current session at the current message
/history In-app scrollback (PgUp / PgDn / Ctrl+↓ also work)
/skills Show injected skills + scores for the current request
/dashboard Compact status overview

Top tools

Every tool is invoked by the model when needed; you don’t call them directly.

Tool What it does
Bash Run shell commands (with safety prompt for destructive ops)
Read Read a file by absolute path
Write Write or overwrite a file
Edit Targeted string-replace edit
Glob Find files by pattern, sorted by mtime
Grep Ripgrep-powered content search
Agent Dispatch a subagent for complex / parallel tasks
Schedule… Cron-style scheduled triggers

Plus 30+ more — see /tools inside the REPL.

Sessions

Every conversation is saved to ~/.owlcoda/sessions/<id>.json automatically. To resume:

owlcoda --resume last           # most recent session
owlcoda --resume <id>           # specific id from `owlcoda sessions`

You can tag sessions (/tag launch-blocker), search by tag (owlcoda sessions --tag launch-blocker), and branch from any message (/branch).

Skills (L2)

OwlCoda matches the user’s incoming message against a curated methodology pack plus any user-installed skills under ~/.owlcoda/skills/. Matched skills get injected into the system prompt at request time. Tune behavior:

{
  "skillInjection": true,
  "skillTopK": 2,
  "skillThreshold": 0.1
}

See owlcoda skills info for what’s loaded.

Training pipeline (L3, opt-in / off by default)

{ "trainingCollection": false }

Or runtime override:

OWLCODA_TRAINING_COLLECTION=1 owlcoda

When on, sessions are scored on 5 weighted dimensions, PII-sanitized, and appended to ~/.owlcoda/training/collected.jsonl. Export:

owlcoda training export jsonl --min-quality 70
owlcoda training export sharegpt --tool-calls-only