@burtson-labs/bandit-stealth-cli 1.7.466 → 1.7.467

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/README.md +32 -12
  2. package/dist/cli.js +1133 -1069
  3. package/package.json +3 -2
package/README.md CHANGED
@@ -179,18 +179,38 @@ Servers spawn lazily on first invocation, persist for the session, and get clean
179
179
 
180
180
  ## Recommended models
181
181
 
182
- Pull one with `ollama pull <model>`. Bandit auto-detects each model's capabilities and takes the native tool-calling path when supported.
183
-
184
- | Model | Where | Notes |
185
- | ----------------------- | ----------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
186
- | `bandit-logic` (cloud) | Bandit gateway (API key) | **Default for cloud.** Agent-tuned wrapper around Qwen 3.6 27B with thinking mode. Best reliability on multi-step agent tasks — what we recommend trying first. |
187
- | `qwen3.6:27b` | Local / Mac 48GB+, high-VRAM GPU (~17 GB) | **Best local pick.** Same family as `bandit-logic`, runs offline. Probes the filesystem instead of asking for clarification — real agent behavior. |
188
- | `gemma4:26b` | Local / Mac 32GB+ (~17 GB) | Solid alternative when Qwen 3.6 is too heavy for your hardware. Multimodal, 128K context. |
189
- | `gemma4:e4b` | Local / laptop-class (~3 GB) | Lightweight pick that punches above its weight. Validated on real Bandit runs — clean tool sequencing (`ls` → narrow → `read_file`), no hallucinated paths. Right pick when you want a local agent that doesn't pin your fans. |
190
- | `gemma4:31b` | Local / Mac 64GB+, GPU node | Bigger context, better reasoning for complex refactors. |
191
- | `qwen2.5-coder:7b` | Local / Mac (~4.7 GB) | Fast lightweight pick. Native tool calling. Best for "given context, do X" tasks rather than autonomous discovery. |
192
- | `devstral:latest` | Local / Mac 32GB+ | Mistral's agent-tuned model — strong tool use. |
193
- | `bandit-core-1` (cloud) | Bandit gateway (API key) | Lightweight cloud option. Faster first-token than `bandit-logic`, less reliable on multi-step agent tasks. |
182
+ Bandit auto-detects each model's capabilities and takes the native tool-calling
183
+ path when supported. All four cloud models emit native tool calls reliably —
184
+ that is not what separates them.
185
+
186
+ ### Cloud — Bandit gateway (API key)
187
+
188
+ | Model | Notes |
189
+ | ---------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
190
+ | `bandit-logic-2` | **Start here.** Fastest (1.3–1.7s warm, and steady under concurrent load), 256K context, and the model that handled a one-prompt "write a PDF with a diagram" end to end. Frontier coding model. |
191
+ | `bandit-core-2` | Highest score on our 26-fixture agent eval (20/26 vs 19/26), but roughly 3× the wall time for the same suite. Reach for it when quality matters more than latency. 256K context. |
192
+ | `bandit-logic` | Self-hosted. Fast in isolation (~6.5s) but latency degrades sharply when more than one session uses it at once. Agent-tuned, thinking mode, 64K context. |
193
+ | `bandit-core-1` | Self-hosted, lightweight. Same concurrency limit as `bandit-logic`, and availability has been uneven — measured between 8s and 112s for the same trivial prompt on the same day. Check before relying on it. |
194
+
195
+ > **Concurrency matters more than the benchmark.** `bandit-logic` and
196
+ > `bandit-core-1` run on our own hardware and serve roughly one request at a
197
+ > time: two simultaneous sessions are enough to stall both. `bandit-logic-2`
198
+ > and `bandit-core-2` are cloud-backed and hold their latency under load. If
199
+ > you are fanning work out in parallel, use the cloud pair.
200
+
201
+ ### Local — Ollama
202
+
203
+ Pull one with `ollama pull <model>`.
204
+
205
+ | Model | Where | Notes |
206
+ | ------------------------ | ----------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
207
+ | `qwen3.6:27b` | Local / Mac 48GB+, high-VRAM GPU (~17 GB) | **Best local pick.** Same family as `bandit-logic`, runs offline. Probes the filesystem instead of asking for clarification — real agent behavior. |
208
+ | `gemma4:26b` | Local / Mac 32GB+ (~17 GB) | Solid alternative when Qwen 3.6 is too heavy for your hardware. Multimodal, 128K context. |
209
+ | `gemma4:e4b` | Local / laptop-class (~3 GB) | Lightweight pick that punches above its weight. Validated on real Bandit runs — clean tool sequencing (`ls` → narrow → `read_file`), no hallucinated paths. Right pick when you want a local agent that doesn't pin your fans. |
210
+ | `gemma4:31b` | Local / Mac 64GB+, GPU node | Bigger context, better reasoning for complex refactors. |
211
+ | `qwen2.5-coder:7b` | Local / Mac (~4.7 GB) | Fast lightweight pick. Native tool calling. Best for "given context, do X" tasks rather than autonomous discovery. |
212
+ | `devstral:latest` | Local / Mac 32GB+ | Mistral's agent-tuned model — strong tool use. |
213
+ | `bandit-core:12b-it-qat` | Local / Mac (~9 GB) | Our Gemma-12B fine-tune. Fits most laptops, agent-aware. |
194
214
 
195
215
  ### Models we don't recommend (for agent work)
196
216