omp-alamode
Aggregate Oh My Pi plugin bundle: lemmeaskchat, tokenmunchers, let-me-see, flymode, and built-in session usage awareness with manual PAYG balance estimates
Package details
Security note
Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.
README
omp-alamode
Four Oh My Pi (omp) plugins in one package, plus optional usage awareness for context, costs, and provider quotas, and a standalone /hurry finish-fast prompt.
Included features
| Feature | Command | Plugin |
|---|---|---|
| Discord questions and away-mode conversations | /lemmeaskchat |
omp-lemmeaskchat |
| Token-usage leaderboard | /munch |
omp-tokenmunchers |
| Live multi-agent dashboard | /agent-tree |
omp-let-me-see |
| Display-only fly-language masking | /flymode |
omp-flymode |
| Context, session usage, quota information, manual credit estimates, and finish-fast steering | /usage-awareness, /hurry |
Built in |
The four upstream plugins keep their original behavior and configuration. This package loads them together; it does not replace their implementations or add a shared settings layer. /hurry belongs to the built-in usage-awareness extension rather than a separate extension entry.
Installation
Requires Bun 1.3.0 or later and omp 18.20.2 or later.
Once the package is published to npm:
omp plugin install omp-alamode
Start a new omp session after installing.
Avoid duplicate loading: uninstall or disable standalone copies of omp-lemmeaskchat, omp-tokenmunchers, omp-let-me-see, and omp-flymode before using this bundle. Do not load them separately with omp -e alongside it; commands and hooks would register twice.
Setup
- Discord relay: requires a Discord bot token and your Discord user ID.
- Leaderboard: requires an account and API key; use
/munch login. - Agent dashboard and fly mode: no credentials required. Fly mode is off until enabled.
- Usage awareness: no separate credentials or setup. When enabled, it uses the host's existing provider configuration.
/hurryneeds no setup and works whether awareness is on or off.
Existing plugin settings are reused and left untouched. See each upstream plugin's README for its full configuration and command reference:
Usage awareness
When enabled, usage awareness gives the model a short usage summary and finish-fast guidance for each request. It is off by default and must be enabled explicitly.
| Command | Action |
|---|---|
/usage-awareness |
Toggle on or off |
/usage-awareness on |
Enable for this session |
/usage-awareness off |
Disable and stop adding usage context |
/usage-awareness status |
Show the state and, when enabled, current usage information |
/usage-awareness balance |
Open a provider picker and enter a manual credit estimate |
/usage-awareness balance <USD> [low-USD] |
Record a manual credit estimate for the current provider without a menu |
/usage-awareness balance clear |
Remove the current provider's manual estimate |
When enabled, the summary can include:
- Context window: tokens used, capacity, and remaining space for the active model.
- Session usage: token counts and host-recorded costs, including subagent costs.
- Provider quotas: available limits, remaining allowance, reset times, and report age.
- Manual credit estimate: an estimated remaining dollar range, if you entered a balance.
These are separate measurements: context capacity is not a provider quota, and session cost is not a provider invoice.
While enabled, each injected usage context also includes a compact finish-fast instruction: complete the current task with only the required work, avoid extra scope, unrelated cleanup, and optional polish, and skip optional tests and reviews unless explicit requirements or higher-priority instructions require them. The instruction does not ask the model to claim that skipped checks passed, does not override higher-priority instructions, and does not guarantee faster completion. It is appended even when usage information cannot be read.
Finish fast with /hurry
/hurry is a one-shot request to finish the current task quickly. It is a standalone command in the same built-in extension, not a mode: there is no persistent setting, and usage awareness does not need to be enabled. Arguments are generic modifiers, not special vocabulary: only the number of whitespace-separated words matters, and the words are never sent to the model. off and status are just modifiers, not modes.
| Command | Level |
|---|---|
/hurry |
1 |
/hurry fr or /hurry ong |
2 |
/hurry fr ong |
3 |
/hurry fr ong please |
3 (capped) |
Each call is independent: levels do not accumulate across calls, and extra modifiers beyond two are capped at level 3. Urgency increases with level: 1 asks to finish promptly with minimal scope; 2 asks to finish urgently, skipping optional investigation and rechecks; 3 asks for only strictly necessary work, then a concise report.
- While a turn is running, the prompt is delivered as steering guidance so the model can adjust the current task.
- When idle, it starts a turn.
- Every level preserves required instructions and asks for no extra tests, reviews, or new subagents unless explicitly required.
- Every level asks the model to report what it verified and what it skipped. It cannot override higher-priority instructions and cannot guarantee faster completion.
/hurry sends an explicit user message, which the host records as normal session history. It makes no provider quota queries and does not change awareness state or credit balances. Unlike the usage summary, which is temporary outgoing context, that user message persists in the transcript.
Privacy and limits
While off, the extension makes no provider quota queries and adds no usage context. Setting or clearing a manual balance while off only changes local state; it does not enable awareness or trigger a provider query.
All extension state is held in memory. Starting, resuming, switching, or branching a session resets awareness to off and clears manual balances. Nothing is saved to disk.
When enabled, quota reads go through the host's public, normalized usage service—not direct provider requests from this extension. Reports up to five minutes old are treated as fresh; older retained reports are historical only. Missing or failed data is unknown, and a past reset time does not prove that allowance has renewed.
Quota reports are account-scoped snapshots, not guaranteed allowance for the account used by the next request. Separate account reports are not combined. Raw reports, errors, credentials, and account identifiers are not added to model context.
The summary, including its finish-fast guidance, is temporary outgoing context, not saved chat history. Only one copy is added per request, and pending results are discarded after disabling. It adds a small amount of request context but never switches models, changes credentials, aborts tasks, or compacts the session.
Manual pay-as-you-go credit estimates
If your provider does not report a usable dollar balance, open the balance menu:
/usage-awareness balance
Choose a provider, enter its remaining USD balance, then enter a low-credit threshold or leave it blank for 10%. The picker lists providers from the host's model catalog, including the active provider; choosing one does not switch models or require an active model. Cancel any dialog to leave balances unchanged.
For the active provider, you can also enter amounts directly:
/usage-awareness balance 25
This records an estimated starting balance of $25, with a default low-credit threshold of $2.50 (10%). To choose a different threshold:
/usage-awareness balance 25 5
Amounts are non-negative USD values. The optional threshold is an absolute dollar amount. Each balance belongs to the provider selected in the menu, or the active model's provider when using the numeric shortcut; switching providers does not transfer it. The numeric shortcut and balance clear require an active provider. Estimates and reminders follow the active provider, so a balance entered for another provider is used when you switch to it.
The extension records spending baselines when you enter the amount, so earlier recorded costs are not deducted again. It then shows two scenarios, each floored at zero:
- Low scenario: starting balance minus provider-labelled spending and all recorded task/subagent spending since entry, assuming those child charges used this provider.
- High scenario: starting balance minus provider-labelled spending only.
Both are scenario estimates, not guaranteed bounds. They cannot certify a budget. Task/subagent charges lack reliable provider attribution. Provider-labelled spending includes recorded assistant and model_usage costs for that provider, including entries on inactive branches. Child spending combines the host's subagent total with task-result details, including nested tasks, without double-counting overlaps.
The estimate excludes outside-session spending, usage without a recorded USD cost, child model_usage, and task/subagent spending absent from task-result details. It is not synchronized with provider billing, quotas, or account balances. Actual remaining credit may be lower; check your provider for the authoritative balance.
Re-enter the provider's current balance to correct the estimate after a top-up, outside charge, or billing drift. Re-entry replaces the amount and both spending baselines; it does not add credit to the previous estimate.
Possible low-credit and exhaustion reminders use the low scenario, appear only while awareness is enabled, and fire once each per balance entry. Re-entry resets them. They are reminders—not spending caps or enforcement.
Compatibility
The host requirement remains @oh-my-pi/pi-coding-agent >=18.20.2. Compatibility was checked against compiled omp 18.22.0 with Bun 1.4.2; no dependency or wrapper changes were needed.
| Bundled extension | Verified version |
|---|---|
omp-lemmeaskchat |
0.3.0 |
omp-tokenmunchers |
0.2.0 |
omp-let-me-see |
0.3.0 |
omp-flymode |
0.1.0 |
| Usage awareness | Built into alamode 0.2.1 |
Verified scope
- Loaded all five extensions together with
omp --no-extensions -e <checkout>in an isolated environment. Each bundled command registered exactly once, with no extension-load errors. - Executed
/lemmeaskchat status,/munch status, and/flymode off,on, thenoffthrough the compiled host's RPC interface. - Opened and closed
/agent-treein the actual interactive terminal; its dashboard rendered successfully. - Executed
/usage-awareness status, numeric balance entry, the provider-picker/input flow, and balance clear while off. Awareness stayed off; instrumented smoke checks observed zero provider usage-report queries and zero fetch calls. No model prompt was sent. - All 52 deterministic built-in tests passed (596 assertions), covering usage knowledge, credit accounting, and extension lifecycle behavior.
The published 18.22.0 host sources retain the registration, event, UI, and package-import surfaces used by the upstream bundles. Sources: pi-coding-agent 18.22.0, pi-tui 18.22.0, pi-utils 18.22.0, and omp-stats 18.22.0. The repository's development host peer resolves to 18.20.2; it was not used as the compiled runtime for these checks.
The built-in extension's public context-usage, session-entry/statistics, normalized usage-report, model-catalog, dialog, and context-injection contracts also match 18.22.0. Usage-report types were checked against pi-ai 18.22.0; enabled provider reads remain untested against live accounts.
Limits
This does not guarantee compatibility with every later host release. omp-let-me-see@0.3.0 still declares an older exact optional host peer, 18.8.6; the runtime check above is separate evidence. Fly mode patches TUI internals, tokenmunchers reads the host's internal stats database, and lemmeaskchat wraps the native ask schema; future host changes can require upstream updates.
Live Discord, leaderboard, and provider endpoints were not contacted. Authenticated relay/sync, live subagent activity, model-output masking, and enabled quota retrieval were not exercised by the local smoke checks. The recorded compiled-host checks also predate /hurry and the enabled finish-fast guidance; neither was exercised there.
Local development
git clone https://github.com/hndrxc/alamode.git
cd alamode
bun install
omp plugin link .
omp plugin link . registers a persistent development link to your checkout. Use it only when you want your local plugin installation to load this directory.
Source TypeScript is published as ESM; there is no build step. Each upstream plugin has its own extension entry under src/. The built-in extension uses usage-awareness.ts, usage-knowledge.ts, and credit-balance.ts, and /hurry is registered by usage-awareness.ts.
Run deterministic built-in extension tests and inspect the publish contents with:
bun test
npm pack --dry-run --json
Tests use fake host interfaces and require no provider credentials. They do not replace a compiled-host activation check or live-service testing.
License
MIT © Carter Hendricks