This release is a pre-release and may not be stable for production use.
Crowe Logic
The agent console for the terminal. One command, a real tool loop, and a model table that names the engine beside every tier.
crowe runs as an interactive console or as a single prompt. It carries its
own tool registry, reads and edits files, runs shell commands under an
autonomy level you set, searches the web, and streams answers with live progress.
Sign in with Crowe ID for hosted access. The free plan offers CroweLM Flash
at twenty turns a day; model availability depends on the account plan or the
provider credentials configured for a self-managed installation. The console
checks for updates once a day; crowe update checks on demand.
Proprietary software, licensed not sold. See LICENSE. Python 3.10 or newer on macOS, Linux or Windows.
0.11.0rc1 preview: steer the agent while it works
This is an opt-in release candidate, not the stable channel. Try it in an isolated environment without replacing your installed stable version:
uvx --from 'crowe-logic==0.11.0rc1' crowe
uvx --from 'crowe-logic==0.11.0rc1' crowe-logic chat
- Type a correction while the workspace is working, then press Enter to queue it.
- Escape, Ctrl+C, or
/stoppauses at a safe boundary; an active operation may finish first. - Ctrl+X returns queued text to the composer for editing.
- Ctrl+G searches models, engines and hosts; Ctrl+B opens tool evidence.
- Inline chat offers numbered receipts through
/inspectand checkpoint-aware/retry. - Saved conversations retain their selected model variant and interruption warnings.
Tool checkpoints are kept in the current process, not restored after a restart. The full-screen workspace supports macOS and Linux terminals; Windows users should use inline chat, and Windows parity has not been verified for this preview.
Install
pip install crowe-logic
This installs two commands, crowe and crowe-logic, which are the same
program; the long name stays so nothing installed before the rename breaks.
An npm wrapper exists for Node shops and runs the same pip install for you:
npm i -g crowe-logic
First run
crowe # the console, on the free tier
crowe login # sign in with Crowe ID; paid tiers appear in /model
crowe run "prompt" # one prompt, one answer, exit
Inside a session the five commands a new user needs are /model, /effort,
/autonomy, /tools and /help. /replay N reruns a previous turn and
/fork N discards later turns before rerunning it, so recovery does not require
re-explaining project context.
For difficult decisions, operators with an OpenAI-compatible local Astra bridge
can opt in with CROWE_LOGIC_ASTRA_REASON=1, then use /reasoning-engine on.
That adds an independent astra_reason opinion only for consequential analysis
and difficult debugging. It is off by default, context is redacted for common
credential forms before it leaves the process, and /reasoning-engine off
disables it for the active session.
Reasoning is hidden by default in CLI, TUI, and PUI. Use /reasoning on for
provider diagnostic reasoning where supported, /reasoning off to hide it,
or bare /reasoning to toggle. PUI keeps its working indicator rather than
printing raw reasoning. CLI dual panes follow the same setting. Token counts
and progress remain available while reasoning is hidden. /transcript also
honors this setting; hiding reasoning does not erase stored diagnostics.
An explicit CROWE_LOGIC_SHOW_REASONING=on (legacy: CROWELM_SHOW_REASONING)
opts in at startup; the session command takes precedence. This is a display
setting, not a change to model effort or a guarantee of fewer reasoning tokens.
Models
/model lists every tier with the engine it runs on, grouped:
| Tier | Runs on | Notes |
|---|---|---|
| CroweLM Auto | routes per turn | The default. Reads the task and picks a family. |
| CroweLM | GPT-5.6 Terra by OpenAI | Where a turn starts. Three effort rungs. |
| CroweLM Zenith | GPT-6 Astra by OpenAI | The hardest end-to-end work. The other half of dual mode. |
| CroweLM Depth | Grok 4.6 by xAI | Long-horizon work over very large contexts. |
| CroweLM Flash | GLM 5.3 Flash by Z.ai | Drafts and tight iteration. Lowest cost per turn. |
| CroweLM Coder | GPT-5.6 Sol by OpenAI | Code in your codebase. |
| CroweLM Vision, Grower, Kernel | Claude Fable 5.1 | Personas: images, cultivation operations, systems work. |
| GPT-5.5, GPT-5.4, DeepSeek V4 Pro, Kimi K2.7 Code, Kimi K3, GLM 5.3 Flash, Claude Sonnet 5, Claude Opus 4.8 | themselves | Frontier picks, offered by name on your own inference keys. No CroweLM identity. |
| CroweLM Mycelium | Crowe Mycelium (Gemma 4 fine-tune) | The free tier. The model Crowe trained. |
| DeepParallel, Nemotron Nano and Super | multi-provider council; NVIDIA NIM | Backstops. |
Pick by number or name (/model zenith, /model 2). /effort normal|high|max
moves within a family. Crowe Logic does not train the frontier engines; what
makes a tier CroweLM is the routing, the grounding in Crowe's knowledge base,
the tools and the orchestration on top. Ask any tier what it runs on and it
answers plainly. The tier names stay put when an engine moves, so your pins
and scripts do not have to.
Sign-in and plans
The CroweLM tiers run through your Crowe ID at id.crowelogic.com. The free
tier has no card and no keys. Pro is $99 a month for the whole table, every
effort rung, dual mode and synthesis, and the desktop and web apps on the same
sign-in. crowe upgrade opens checkout; crowe billing opens the customer
portal; crowe whoami shows the signed-in identity and tier.
The frontier picks and any self-hosted endpoint run on keys you hold:
crowe init --node writes the env template.
Usage
crowe # interactive console (default)
crowe run "prompt" # one prompt, one answer, exit
crowe login # sign in with Crowe ID
crowe whoami # signed-in identity and tier
crowe upgrade # open Pro checkout
crowe billing # open the customer portal
crowe doctor # check the install and its connections
crowe tools # list the tool registry
crowe update # update now (the console also does this daily)
Bring your own keys
The CroweLM tiers need only your Crowe ID. The frontier picks, and any
endpoint you host yourself, run on keys you hold. crowe init --node writes
~/.crowe-logic.env with the variables below; fill in the ones you use.
| Variable | Purpose |
|---|---|
CROWE_LOGIC_GATEWAY_URL |
The Crowe Logic gateway; defaults to https://api.crowelogic.com |
CROWE_ID_ISSUER |
The Crowe ID issuer; defaults to https://id.crowelogic.com/realms/crowe |
CROWE_LOGIC_AUTO_ROUTE |
1 lets CroweLM Auto resolve to a live tier per turn on a self-managed node |
AZURE_CORE_ENDPOINT, AZURE_CORE_API_KEY |
Your own Azure inference endpoint and key |
CROWE_OPEN_ENDPOINT, CROWE_OPEN_API_KEY |
Any OpenAI-compatible endpoint you host |
NVIDIA_NIM_ENDPOINT, NVIDIA_API_KEY |
NVIDIA NIM, for the Nemotron backstops |
CROWE_LOGIC_HOME |
Where the console keeps per-user state; defaults to ~/.crowe-logic |
CROWE_LOGIC_TELEMETRY |
0 turns off the local telemetry file described below |
CROWE_LOGIC_ASTRA_REASON |
1 opts into the independent Astra bridge; off by default |
CROWE_LOGIC_ASTRA_REASON_URL, CROWE_LOGIC_ASTRA_REASON_MODEL |
OpenAI-compatible bridge base URL and model selector |
Extra model entries load from CROWE_LOGIC_EXTRA_MODELS_JSON,
CROWE_LOGIC_EXTRA_MODELS_PATH, or ~/.config/crowe-logic/models.extra.json.
MCP server
The Crowe Logic platform is also available as a standalone MCP server for any MCP client:
uvx crowe-logic-mcp
The bundled crowe-logic-quantum-mcp server needs the mcp extra:
pip install "crowe-logic[mcp]"
Specialized agents
| Agent | Focus |
|---|---|
cultivation |
Mushroom cultivation with the Crowe knowledge base |
quantum |
Quantum computing with QubitFlow + Synapse + Trinity bridge |
music |
Music composition with the Talon engine |
code |
General software development |
research |
Web research and analysis |
Local telemetry
Crowe keeps a local record of what the console did: which tools ran and how long
they took, which tier and provider answered, token counts, timings, and session
events. It does not keep your prompts, the replies, or the contents of your files.
Tool arguments are cut to 200 characters, and the values of content, text, and
data arguments are dropped, with only their lengths recorded.
The record is a JSON-lines file at ~/.crowe-logic/runtime/telemetry.jsonl, created
with owner-only permissions and rotated at 50 MB. It stays on your machine; the
console never uploads it. Nothing is written until the first event, so
crowe --version touches nothing under your home directory. To turn it off, set
CROWE_LOGIC_TELEMETRY=0 (false, no, and off also work).
Requirements
- Python 3.10 or newer on macOS, Linux or Windows.
- A Crowe ID for the paid tiers; nothing for the free tier.
Author
Michael Crowe, Crowe Logic, Inc.
License
Proprietary. Copyright (c) 2026 Crowe Logic, Inc. All rights reserved. See the LICENSE file for terms. Licensed, not sold: use requires the no-cost tier or a current paid plan, and redistribution is not permitted.
Releases before 0.7.1 shipped an Apache License 2.0 text and were distributed under it; this change is not retroactive.
Licensing questions: michael@crowelogic.com
Metadata
Release files for crowe-logic 0.11.0rc5
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| crowe_logic-0.11.0rc5.tar.gz | 2.3 MB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| crowe_logic-0.11.0rc5-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 4.7 MB
Release files / crowe_logic-0.11.0rc5.tar.gz
| Download URL | crowe_logic-0.11.0rc5.tar.gz |
|---|---|
| Size | 2.3 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
ee2495792844be6030dbf04c24c37b1f512facdca06eb36eedfe9fd2443015b2
|
|
BLAKE2b-256 checksum How to use checksums |
5936aca0ccdf7a8a780db4c83479717231f5b6f085e6c6727dd2fb5babb36120
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 18, 2026.
Transparency logRelease files / crowe_logic-0.11.0rc5-py3-none-any.whl
| Download URL | crowe_logic-0.11.0rc5-py3-none-any.whl |
|---|---|
| Size | 2.4 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
416ea4e406e49378577a5e1f716414612584a8b82950273f0287b14b2e344492
|
|
BLAKE2b-256 checksum How to use checksums |
6b6caeecffe12ae5afebf6a88b7ab0a5c278be6212f46a8936ac584644926534
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 18, 2026.
Transparency log