Three corrections to the row added in 5396e33b, none of which changes where it is offered:
- Its peak. The run gave the process footprint's LIFETIME maximum, read at the end: 3.0 to
3.3 GB above what was held then. That is a floor, because a larger transient during the
earlier, longer prompts would not show in it. The row now takes the first measured row's
5 GB: 36.1 GB at 65,536 tokens (0.960 of a 48 GB Mac's estimate) and 39.0 GB at 131,072
(0.972 of what that Mac's runtime measures).
- What was compared. The two builds were not given the same prompt in OpenCode: the GGUF
build got a one-line task through the skill's sendmark (which adds its answer-file
sentence), the MLX build a plain one-line prompt. And the speeds to set side by side are
the two from the same 300-token request, 15.2 and 30.2.
- Runtime issue 18269: whether the memory fix its maintainer promised predates the pinned
runtime is not known, and the header says so.
Also a stale comment: the catalogue holds eleven models.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Run on a MacBook Pro (M5 Pro, 48 GB, runtime 0.40.15) on 2026-11-12, after the GGUF build
this branch lists turned out to be unusable in agent use there: an OpenCode turn of one
line showed no answer after 7 minutes, twice, although the runtime alone wrote 15 tokens
per second.
The MLX build of the same model (`muse-glimmer:30b-mlx`, 19.1 GB) on that Mac:
- 32.6 tokens per second for a 1,500-token answer. The tag carries a draft model, and the
runtime logs speculative decoding with 0.34 to 0.62 of the drafts accepted.
- A correct tool call in under 3 seconds, and the same OpenCode turn finished in 45 seconds.
- Memory, read from the runner's own log line after every request: 37,900 to 45,100 bytes
held per token (the GGUF build computes to 7,072), prefix snapshots that filled their
8 GiB during twelve separate prompts, and a process peak 3.0 to 3.3 GB above what stayed
held.
So it gets a row of its own, the second with measured figures. They put it at 34.6 GB for
65,536 tokens: offered from 48 GB up, never on the 32 and 36 GB Macs, which keep the GGUF
row untouched (it is still the pick of a 32 GB Mac with a Max chip: whether it should be
is the maintainer's decision). It is never the pick of a Mac whose disk has room for
Qwen3.6 35B-A3B, and it has no `tested` entry: one turn is not a test.
Also written down from the same session: the session of many prompts the Qwen3.6 35B-A3B
row owed (5.6 GiB of prefix snapshots after twelve prompts, still growing; its figures are
unchanged), and what the two open runtime issues on the MLX tag say against what ran.
Not done: the row was not installed through the window, and the measurement used a second
instance of the pinned runtime, not Codeman's.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
A 16 GB Mac got no model: the two 9B builds of the catalogue need 10.2 GB at
65,536 tokens, 0.987 of the estimated budget, just past the 0.98 a tight fit
may reach.
New row `ornith:9b-q4_k_m` (Ornith 1.0 9B, MIT), a fine-tune of Qwen3.5-9B for
coding agents whose GGUF build has no vision projector: 5.63 GB against 6.55 GB,
so 9.3 GB at 65,536 tokens, 0.898 of that budget. SWE-bench Verified 69.4,
SWE-bench Pro 42.9, Terminal-Bench 2.1 43.1 on Ornith's card. Its config.json
and its GGUF header are those of Qwen3.5-9B (no prediction layer), so it shares
the figures of the two other 9B rows. Never run on a Mac: `estimated`.
- What changes for a Mac: 16 GB goes from no model to `limited` with this row,
tight. That is the only pick that changes: the row is ranked under MiMo V2.6
9B, so every Mac that could already run a 9B keeps its pick and gains an
alternative. With the budget a 16 GB Mac mini measured (10.7 GB) the pick is
MiMo as before.
- Why under MiMo although its own card's numbers are higher: the two vendors'
tables put their common base 7 points apart, and runtime issue 16992 (open)
was filed on the 35B sibling, which has the same renderer and parser. The
35B stays set aside.
- The 16 GB rule itself is untouched: same budget estimate, same 0.98, same
eight context checkpoints.
- Ranks renumbered 1 to 10; the pins of test/local-model-recommend.test.ts and
the table of docs/local-models.md follow.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The "Run a local model" chooser showed the pick and four more models. It now
lists every catalogue model that fits the Mac, and the install accepts every
listed row.
- MAX_ALTERNATIVES goes from 4 to 11, above what the catalogue holds. It stays
a bound: test/local-model-catalog.test.ts fails for a catalogue that outgrows
it, so nothing drops off the list unseen.
- New row `muse-glimmer:30b-q4_k_m` (Meta, Apache-2.0): SWE-bench Verified 76.0,
SWE-bench Pro 51.2, Terminal-Bench 2.1 51.7 on Meta's card. Its K/V cache is
the smallest of the 27B class, so it is the first model of that class a 32 GB
Mac holds: 21.2 GB at 65,536 tokens, 0.970 of the estimated budget. Figures
computed from its config.json and its GGUF header by the llamacpp rule (13
full-attention layers, 39 sliding-window layers of 2,048 tokens, no
prediction layer); the sliding-window cache rule was read again in llama.cpp
b12111 (src/llama-kv-cache-iswa.cpp). Never run on a Mac: `estimated`.
- What changes for a Mac: 32 GB is no longer `limited`. On an M1 Max or M2 Max
the pick is Muse Glimmer; on a base or Pro chip the dense model is slow and
the pick stays MiMo V2.6 9B, now with the small-pick note. Every other pick
is unchanged.
- The "set aside" list was read again at its sources on 2026-11-10 and gained
ripple-preview:65b (every build requires runtime 0.40.16, one above the pin),
Ornith 1.0 (runtime issue 16992), nemotron-3.5-super:120b, qwen3.5:4b,
qwen3-coder-next and the MLX and DFlash tags of Muse Glimmer.
- The header and the guide now say that runtime issue 16992 has one diagnosed
cause in the tool parser every Qwen-family row shares (a tool argument that
quotes a whole tool call is cut off; the fix is runtime PR 17021, open).
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The install window polls GET /api/local-model every second. A read of the
state file that was still out while the job wrote its last state and let
go of this.job came back with the old `installing`, and settling that with
no job in the process read it as an interrupted job: `paused` with Resume
for one tick before `failed`. Under CI load this made
test/local-model-install-e2e.test.ts fail on a different failure case per
run (master 86502b69, PR #790).
currentState() now reads again when the job ended during the read. The job
writes before it lets go, so the second read has the final state.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The previous commit counted 4 rollback copies of the recurrent state for every GGUF with a
prediction layer, the runtime's default `draft_num_predict`. A tag can set its own in its
params layer, and the runtime then passes that as `--spec-draft-n-max`, which is what
`need_n_rs_seq()` turns into rollback copies. Read on 2026-11-08:
- qwen3.6:35b-a3b-mtp-q4_k_m: draft_num_predict 2, so 2 copies (was counted as 4).
- qwen3.8:27b-mtp-q4_k_m: draft_num_predict 4.
- qwen3.8:27b-q8_0: none in its params, so the runtime's 4.
The 35B-A3B GGUF row's overhead drops by 132 MB: 0.971 instead of 0.976 on a 36 GB M3 Pro or
M3 Max, the same pick. test/local-model-catalog.test.ts now carries the drafted tokens per row
instead of a yes/no prediction layer.
Also states what was verified about llama.cpp's checkpoint default: 8 from b6800 to b8200,
32 from b8250 together with a checkpoint every 8,192 prompt tokens (common/common.h at those
tags), in place of an inferred reason. Checked in the b12111 source as well: the draft context
of a Qwen-family MTP model caches the prediction block alone (`mtp_on_hybrid_qwen`, layers at
or above n_layer()), and the main context leaves it out, so the row counts that layer once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The catalogue that names each row's runner (feat/local-model-catalog-v2: `runner` is `mlx` or
`llamacpp`, and a row's memory figures follow that runner's rule) needs the weekly check to
notice when a tag starts serving a Mac another runner: a tag that gains an mlx child, or a
list that loses it. The size check usually catches that, but not when the two builds come out
about the same size, and the figures are wrong either way.
classifyModel() now raises a `runner` finding when the row names a runner and the build a Mac
selects runs on another one. catalogueRunner() maps the registry's names into the catalogue's:
`mlx` stays `mlx`; `llamacpp`, and `ggml` (what Ollama calls a GGUF manifest that names no
runner), are `llamacpp`, because Ollama 0.40.15 serves every GGUF build through llama-server.
A catalogue without the field (the one on this branch today) is never held to a runner.
Run live on 2026-11-08 against the v2 catalogue's eight rows: exit 0, every runner matches.
Tests: the mapping, the finding (both directions, ggml, an unknown runner, a row without the
field), the report text, and runCheck against the fake registry with a plain GGUF manifest.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The llama.cpp memory rule of the catalogue left out two allocations that the llama.cpp the
runtime bundles (b12111, the runtime's LLAMA_CPP_VERSION) makes for every GGUF row:
- Context checkpoints. For a model whose state cannot be rolled back (the recurrent state of
the Qwen-family rows, the sliding window of North Mini Code), llama-server keeps host copies
of that state, two or three per request, up to 32 per slot by default
(tools/server/server-context.cpp create_checkpoint). For the 27B that is 32 x 157 MB = 5 GB,
none of which was counted. The runtime now sets LLAMA_ARG_CTX_CHECKPOINTS=8 (read by
llama-server from its environment like LLAMA_ARG_CACHE_RAM; 8 was llama.cpp's own default
before it went to 32), and every llamacpp row counts 8 copies of its state.
- Multi-token prediction. The two -mtp tags and qwen3.8:27b-q8_0 carry a prediction layer
(nextn_predict_layers 1, read from the first bytes of each GGUF; the 9B tags and North Mini
Code have none). The runtime then always passes --spec-type draft-mtp --spec-draft-n-max 4,
so llama.cpp keeps 4 rollback copies of the recurrent state (need_n_rs_seq(),
llama-memory-recurrent.cpp) and the prediction layer's cache in the draft context's cache
type, f16 by default. The rows counted one state and that layer at q8_0.
What moves (test/local-model-recommend.test.ts, "every memory size", re-pinned):
- 16 GB: limited -> none. MiMo 9B at 65,536 tokens needs 10.2 GB, 0.987 of the 10.4 GB
estimate. With the budget the Mac mini measured (10.7 GB after the reserve) it is offered,
0.954, tight; that test is kept. With 4 checkpoints instead of 8 it would be 0.966 on the
estimate: this size sits on the edge, and only a run on such a Mac settles it.
- 18 GB: the same pick, 0.964 instead of 0.928.
- 36 GB: the same picks, 0.966 (27B on M4/M5 Max) and 0.976 (35B-A3B GGUF on M3).
- 48 GB on a Max chip: the 27B at 262,144 tokens is tight (0.853) instead of good.
- 32 GB with a measured budget of 0.74: North Mini Code at 131,072 tokens (was 262,144), the
27B at 65,536 on a Max chip (was 131,072).
- Every MLX row, and everything from 64 GB up, is unchanged.
test/local-model-catalog.test.ts recomputes the new rule from the architecture table and holds
the catalogue to LLAMA_ARG_CTX_CHECKPOINTS. The catalogue header, runtime.ts, the types,
docs/local-models.md, CLAUDE.md and architecture-invariants say the same. Read in the source,
not measured.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The new GGUF rows carry lowercase quantization tags (`-q4_k_m`); pin that
the launch path's model id test (`cliAcceptsCustomModelId()`) accepts every
stock id for every harness the catalogue offers, so a row the chooser shows
can always be started.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The catalogue had seven Qwen-family rows that all carried the memory
figures measured on one model (Qwen3.6 35B-A3B on the MLX runner), so a
9B or a dense 27B was judged with figures that do not describe it, two
rows had the wrong Mac download size, and every Mac below 32 GB got
"none".
Now eight rows, each with the runner a Mac really uses for its tag (a new
`runner` field: `mlx` or `llamacpp`) and memory figures of its own:
- the 35B-A3B MLX row keeps its measured figures and the tested pair;
- every llama.cpp row is computed from its model's config.json: a q8_0
K/V cache for the whole context (what the runtime now sets), its
recurrent or sliding-window state, 1.0 GB of runner and the 1 GiB host
prompt cache;
- the one other MLX row (Ornith 1.5 122B) is computed the same way,
scaled by what the measured row showed, plus the runner's 8 GiB of
prefix snapshots.
Sizes are the registry's Mac pull totals (2026-11-08, lowercase ids
verified there). What each memory size gets: 8 GB none; 16 to 32 GB a 9B
(limited); 36 GB Qwen3.8 27B or the 35B-A3B as GGUF; 48 GB the tested
pair (or the 27B on a Max chip); 64 and 96 GB the 27B or the 35B-A3B;
128 to 512 GB Ornith 1.5 122B. Nothing but the one row was measured.
The catalogue test recomputes every computed figure from a table of
architectures and lets two rows share figures only when they share the
architecture and the runner. The rule tests of recommend() now run on a
frozen copy of the 2026-11-06 catalogue, so a weekly catalogue update
changes only the per-memory table and the real-Mac tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
runtimeEnv() now also sets OLLAMA_KV_CACHE_TYPE=q8_0, OLLAMA_FLASH_ATTENTION=1
and LLAMA_ARG_CACHE_RAM=1024. Read in the v0.40.15 source of the pinned
runtime: the first becomes --cache-type-k/-v of llama-server (17/32 of the
f16 cache), the second --flash-attn on (a quantized V cache needs it, and
"auto" or "off" would fail such a load), and llama-server inherits the third
from ollama serve's environment, which bounds its host prompt cache to
1 GiB instead of letting it grow by gigabytes in a session. The MLX runner
reads none of them.
The catalogue's llama.cpp rows are computed for these values, so the two
change together. Not yet measured on a Mac.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every App Settings Save up to 1.49.0 wrote `agentSkillEnabled: false`, chosen
or not (the switch loaded an absent key as off and Save sends the whole form),
so the new default never reached an install that had ever saved. On its first
start the server now replaces a stored false with true, once per data dir, and
logs one line when it does. It cannot tell a deliberate off from the old
default; that is accepted and stated in the release notes. An off set after it
ran stays off.
It stores true rather than deleting the key: the page's merge overwrites only
the synced keys the server has, so every device that saved on 1.49 kept its own
false, showed the switch off and wrote false back on its next Save (measured in
the JSDOM harness). True wins on every device, including a browser still on a
cached settings-ui.js. For a server with no key the page now drops the device's
copy as well, so the switch shows the default.
The "done" marker is its own file, `agent-skill-default-migrated.json` in the
data dir: it is not a setting, so GET /api/settings never returns it, no Save
can send it back (the strict schema would refuse it) or drop it (which would
re-arm the migration), and a config-backup restore leaves it alone. It is
checked before anything is read and written only after the settings write
landed. The migration runs in WebServer.start() before the first settings read
and before listen (skipped in test mode, like the other boot-time writes),
writes only through mutateSettingsFile(), drops the 2 s settings cache, and
leaves a settings.json that does not parse alone with no marker, so the next
start after the fix runs it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This reverts commit 0aa34da8, which made the codeman agent skill default on here as well. That change lives in feat/agent-skill-default-on, together with the self-ignoring copy and the reference-folder symlink check it needs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
session.test.ts created a session in the fixed `/tmp/test`, so with the agent
skill on by default every `npm test` wrote a skill copy into
`/tmp/test/.claude/skills/codeman/` and left it there. It now uses a temp
folder of its own and removes it afterwards.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Measured on opencode 1.3.0 in a private tmux socket, with a fake
OpenAI-compatible endpoint streaming the reply a character or two at a
time: the marker was on the screen, but the stripped client stream read
D, spinner, O, spinner, N, ... because opencode redraws its spinner
between the streamed pieces, so wait-output never matched it. sendmark
then waited its whole timeout and answered 2 for a reply that had
finished, with the answer file already written (opencode's write tool
wrote it without asking).
The marker wait now runs in rounds of at most 10 s, and between rounds
the answer file is a way out too: a non-empty regular file whose
checksum held still for one round is read, with matched false. The
marker stays the fast path. The answer-file wait after a marker no
longer spins when the exit wait gets no answer, and treats an ended
wait as the worker gone.
The same run measured the readiness mark: with a fresh home and a
custom provider, the first composer bar came with the composer and a
prompt pasted with its Enter at that moment was submitted, so the docs
say so instead of "not measured". The example keeps its input in a
subfolder, apart from the case's CLAUDE.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Five gate files created sessions with `workingDir: '/tmp'` against a real
server. With the agent skill on by default, every `npm test` then wrote a
skill copy into `/tmp/.claude/skills/codeman/`, outside the per-file temp
HOME, where a Claude Code started in `/tmp` would load it (the workspace
hooks already landed in `/tmp/.claude/settings.local.json` the same way).
Each file now uses a temp folder of its own and removes it afterwards.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Install and uninstall checked only the copy and its parent for symlinks, so a
`reference` folder that is a link carried every write and unlink of
`reference/<name>.md` to wherever it pointed. Measured: install overwrote a
`history.md` outside the workspace and uninstall deleted it. A cloned
repository can commit that link, and a Docker case's container can plant it
in the bind-mounted workspace the host injects into. With the skill on by
default, both now ran without the user asking.
Both paths now return 'symlink' for a linked folder inside the copy, before
the first write or unlink, and the refusal messages say so.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
agentSkillEnabled now reads an absent key as ON, the way
workspaceHooksEnabled does: only an explicit false turns it off. A new
session therefore finds the codeman skill in its folder without a
setting first, and with it the verbs that hand work to other CLIs and to
the local model on this Mac. Settings shows the switch on for the same
absent key.
The injection itself is unchanged: add-only, marker-owned, never through
a symlink, never for remote sessions or a plain shell. Turning the
setting off still removes nothing already written.
The quick-start test now pins the new default (injected on an absent
key, nothing after an explicit false), and the docs that said "default
off" say when that changed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A row can pin its build only as `digest`, the value /api/tags reports
after the pull, and LocalModelEntrySchema must gain that optional field
before the weekly check can compare it. The script header and the docs
now say so, so the field cannot arrive under another name and be ignored.
Until it exists, the report's digest table names the build a Mac pulls
for each row and asks for the rows with measured figures first: their
memory figures describe one build, and without a digest a rebuild of
about the same size goes unnoticed. A switch between the mlx and the gguf
build is still caught by the size check (2% or more apart).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
scripts/check-local-model-catalog.mjs reads every catalogue tag the way
Ollama 0.40.15 on Apple silicon reads it: it asks for the manifest list,
takes the mlx build when the tag offers one (else llamacpp, else ggml),
and compares that build's pull total (layers plus config) with the row's
`bytes`. It also reports a tag that is gone, a build whose `requires` is
newer than the pinned runtime, a pinned digest that moved, a newer Ollama
release, and a pinned ollama-darwin.tgz that is gone or changed. It never
requests a layer, bounds every request, retries nothing and validates
everything that reaches the report. Exit 0 clean, 1 drift, 2 failed.
The plain Docker manifest of a list tag is the llamacpp build, not what a
Mac pulls. Read the right way, the first run finds two rows sized from the
wrong build: qwen3.5:9b (a Mac downloads 7,639,229,431 bytes, +16.6%) and
qwen3.8:27b (18,174,721,847, +2.4%).
.github/workflows/local-model-catalog.yml runs it Mondays at 06:00 UTC
and on dispatch. The check job installs with --ignore-scripts and holds a
read-only token; a separate job with issues: write and no checkout opens
or updates the one "Local model catalogue: weekly check" issue on drift,
comments on it on a clean week when it is open, and is skipped when the
check failed. Actions are pinned by commit.
docs/local-models.md gains "Keeping the catalogue current".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The codeman skill's preamble learns the model Codeman installs on this Mac:
- local_model asks GET /api/v1/local-model and prints one JSON line: the
installed model, the coding agent that runs it and its endpoint, or the
reason there is none (not a Mac, nothing installed, an unfinished
install, a blocked install, admin only, a server without the routes).
Exit 0 only when a worker can start now or after a runtime start.
- spawn_worker <name> local (and name:local in spawn_workers) starts a
stopped runtime the way the window's Start button does, quick-starts the
installed agent with customModel, and waits for its composer (OpenCode's
bar, the claude path for a claude agent, LOCAL_READY_MARK otherwise). A
confirmation question from quick-start fails the spawn with empty stdout;
nothing is ever confirmed on the person's behalf.
- sendmark <sid> <prompt> drives any agent worker without an end-of-turn
signal: the exactly-once input body, a split marker through wait-output,
then the answer read back from a file the worker wrote in its folder
(regular files only, 256 KB at most), with one exit code per outcome.
The preamble stamp moves to 1.50.0 so every cached copy is rewritten.
The tests run the real preamble against a stubbed server.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An absent `agentSkillEnabled` now means on, on the server
(`getAgentSkillEnabled()` reads `!== false`, like `workspaceHooksEnabled`)
and on the page (the App Settings switch loads `!== false`, so a Save of any
other setting no longer writes the old default back). Only an explicit false
turns it off. A partial settings PUT that omits the key leaves it as stored,
since the route merges the request into the file.
The switch's help text says it is on by default, that the copy stays out of
git, and that turning it off removes no copy already written. The shared
route-test mock now defaults to on too, so route tests see what a user sees.
Tests: quick-start against a real server (absent key injects, a partial PUT
keeps it on, an explicit false stays off through a later partial PUT), the
real openAppSettings/saveAppSettings in JSDOM (absent shows on, explicit false
shows off, a Save on a never-touched install sends true and passes the
schema), and the plain create route at the mock's default.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every Codeman-managed copy of the agent skill (`.claude/skills/codeman/`,
`.agents/skills/codeman/`, and the user-level copy) now carries a `.gitignore`
of `*`, written once with O_EXCL before any skill file, the way the prompt
uploads folder does. A user's repository no longer lists the copy as
untracked, and Codeman never touches the repository's own ignore files.
A file already at that name is never overwritten, a copy injected before this
gains the file on its next install (reported as refreshed), and uninstall
removes it only when it is the last thing left and holds exactly `*`, so a
user's extra files in the folder stay ignored. Claude Code 2.1.319 runs
`git check-ignore` only on directories at or above `.claude/skills`, never on
a skill folder, so the skill still loads.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The two stubs from #776 cover the global ignore file and the global config. A core.excludesFile in the system gitconfig still reached the code under test and failed the same test (reproduced with GIT_CONFIG_SYSTEM pointing at one). GIT_CONFIG_NOSYSTEM closes that, as test/git-hooks.test.ts already does for its code under test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflicts resolved as on test/desktop-macbook: config/test-suites.ts keeps both browser tests (desktop-connect-page and local-model-ui), and the CLAUDE.md state-files line keeps #778's cli-installed entry with the local model's local-model.json and local-model/ entries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflicts resolved as on test/desktop-macbook: install.sh usage() lists CODEMAN_NO_ADMIN=1 or 0 and #773's CODEMAN_AI_CLI and CODEMAN_PROGRESS=1; probe.rs keeps #781's agent() with TIMEOUT, MAX_BODY and probe_within all pub; the desktop README keeps both known-limits entries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`GET /api/approvals` re-reads the pane of every pending dialog and resolves
one whose frame shows no dialog, either because its options vanished or
because the frame shows a turn running. `notePrompt` deliberately waits 3 s
before making that same judgement, since the Notification hook can land
before Ink paints the dialog and the working line is still on screen then.
The list did not wait. That rarely mattered while it ran once per page load,
but a client that polls it would resolve a live prompt as answered in the
terminal, clearing the red alert on every device and the push Approve button
with it. `sweepForList` skips a dialog younger than the delay.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When a Codeman server of this user ran at an address the app could not open
as its default one, the This Mac tab only named host and port and sent the
user to Another machine to type the address again. A second instance beside
the one that answers (a beta, a `codeman web --port 4000` in a terminal) was
not shown at all.
The look at this Mac now works out an address for every listening socket of
this user's Codeman servers (`found_addresses`, pure): 127.0.0.1 for a bind
to loopback or to every interface, `[::1]` when that is the only loopback
socket, the address itself for a bind to one; `https` and the base path come
from `--https` and `--base-url` on the command line. A server at the default
address with no `--base-url` of its own keeps the default URL, since
install.sh's launchd job has the base path in its environment, where `ps`
does not show it. The default address comes first, then the others by port.
Each address other than the default one is asked once, all at the same time,
with no login. The default probe's answer is reused for its own row. When
nothing answers at the default address, the first found server that answers
is still what Open is for, so each gets the probe's usual 5 s; next to a
Codeman that answers they are only listed and get 2 s. `LocalStatus.found`
replaces `elsewhere`.
The tab lists them: "Codeman is already running on this Mac." with a row
each when none answers at the default address, and "Also running on this
Mac" under Open otherwise. A row has the address, the version and Connect,
or the reason the app cannot open it. Connect fills the address in and runs
the page's one connect(), so the app checks it, uses its saved login and
remembers it like an address typed under Another machine. Nothing connects
by itself, and a refused connect from a row never arms the reopen that Open
has for a starting server.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The violet button was the one thing on the start screen outside the app's
blue, and it carried white text and a see-through disc where its neighbour
has dark ink and a solid disc. It is now white with dark ink and a solid
disc, the same on the Install button, and the chooser's selected row, tile
and model bar follow. Light skins turn it over to dark, since a white button
vanishes on a pale page. The context takes the app's blue.
The tokens are named for what they mean now (--lm-brand, --lm-context).
Brings test/local-model-ui.browser.test.ts up to the redesign it had fallen
behind: it is outside the gate, so the button's 46 px, the 12 px bar and the
memory panel's edges had gone unnoticed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The owner's verdict on the chooser: nice, but not nice enough, too much just
blue. He also wants "Run a local model" on top of the start screen, above
Run Claude Code.
The memory bar is now a map: the model, its context and the brief peaks side
by side on the track of what this Mac can give a model, each with its size in
a legend, and the verdict on the fit beside the figure. recommend() sends what
the need is made of as `Candidate.memory`.
Colour now carries meaning: green for good and tested, amber for tight, orange
for slow, teal for the context. Violet is the feature's own colour, on its
start screen button and its main action. Every row has a size tile and a speed
gauge scaled to the fastest row in the list. The light skins get deeper values
of the same colours, and the window is 40 px taller to hold it without
scrolling.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The window is one frame for the whole flow: a fixed height on desktop, App
Settings' boxed close button, the buttons in a foot of their own outside
the part that scrolls, so a click never moves the next button away and
nothing is cut off; the short screens are centred in it on desktop and sit
above the buttons on a phone.
The chooser lists the pick first and then the other models by download
size, largest first, a column the reader can see. A pick that only just
fits says "Best fit, tight". The memory figure is an estimate and says so
once, in whole GB; its legend names the peak, and the installed line uses
the same words beside what the model takes now, so the two figures are not
read as a contradiction. The legend, the column headers and the foot's
promise are set in --text-dim, the bar's track in --text-muted so its free
part can be seen, and only the download figure keeps tabular figures.
The consent screen is the download as a table: the chosen row, the runtime,
the coding agent with its installer's command as the row's second line, the
total in the strong weight, then where it goes in one line. The download bar
means the model download alone and is gone once the model is in; the screen
that ends an install shows "Start a session" alone; an error on an installed
model is the headline with a grey dot; an error's first sentence is the
headline; Remove is quiet everywhere but red in its own question. The
no-recommendation reasons are one sentence each, and the untested note is
in Codeman's voice. A refusal the server names is printed from its sentence.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Only the two session routes asked launchRefusalMessage(). The attach that
finds no pane, the respawn of a dead pane (a click on the tab, a respawn
start, the boot recovery), restartCli() and the reboot-restore rebuild
started an agent on the endpoint unasked. Session gains setLaunchGuard(),
asked right before a NEW agent process is started and never for an attach to
an agent that runs; WebServer.setupSessionListeners() sets it for every
session; a refusal is a LaunchRefusedError that never falls back to a direct
PTY. The reboot-restore route asks before it builds and skips the entry with
its own reason, left on offer.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The connect step records the connection in the state file BEFORE it writes
the endpoint record and the settings flag (the flag's earlier value read by
inspectLocalEndpoint()), so a state write that fails or a server killed
between the two never leaves a record of ours that the state file does not
know; a record under the reserved id that says exactly what ours would say
is ours. changePort() writes the state first and puts the old port back when
the endpoint move is refused.
A job whose end could not be written (a full disk) is kept in memory: the
view shows it, Resume and a new install start from it, the next verb writes
it, and Cancel never answers a 500. A first install on a full disk is a 507
with the sentence. "installing" in the file while this process runs no job
always reads as paused or that failure.
An installed model whose runtime does not start says so (runtime-start-failed
with Start as the next step, and the launch check says it too); the stored
runtime version is started while its folder is unpacked. Stop, Remove and
the boot of anything that is not a running installation end a runtime an
earlier server left behind before any file is deleted. Stop and "Use another
port" answer 409 sessions-open while a live session uses the model. A pull
that failed because the runtime is gone is runtime-stopped, not a network
problem. After "does not fit" only a model that needs less memory is offered.
A replaced job's partial blobs are deleted. Every 400 carries its sentence as
data.message, and the six inline sentences are labels with zh-CN entries.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The supervisor reports the child's pid the moment it is spawned, before it
answers, so a server killed during a start still leaves a pid that names the
runtime as ours. stop() takes that stored pid (previousPid) and ends a
runtime an earlier server left behind when its command line is a binary
under our runtime folder, so no caller deletes the folder under a live
process. start() takes a version: an installed model keeps the runtime
version it was installed with when a later build pins another one.
removeLocalModelFiles() refuses a local-model folder that is itself a
symbolic link (every check compared against the real path of the root, so
a link in its place passed whatever it pointed at). removePartialModelBlobs()
deletes what an interrupted download left in the model store, which the
runtime answers "not found" for and never deletes.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The real routes, service, recommendation, runtime download and unpack,
supervisor, pull, checks, connection and state file, joined together. Only
the runtime (a fake HTTP server the injected spawn "starts"), the hardware,
the coding agent's installer and the waits between retries are stand-ins.
Walks: the whole install and its removal (the folder is gone, the settings
flag is restored); a runtime download that arrives damaged; a model download
that fails twice and then succeeds; one that keeps failing; a model that
calls no tool; Cancel during the model download; a server restart during the
model download, which comes back paused and waits for a click.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The read-only card is gone. The welcome screen has one button, "Run a local
model" ("Local model" once one is installed), and App Settings > Models has
one row; both open the same window.
The window follows `install.state`: choose a model (a radio group, with a
memory bar filled to what the selected model needs), confirm the sizes and
the folder, watch the download and the steps, and end on "Start a session".
Paused, failed, installed (running or stopped), Remove and "no model fits"
are screens of the same window. It polls the GET once a second while open.
Every dynamic string goes in through textContent, labels and sentences come
from the payload, no CLI id is compared, and every new string has a zh-CN
entry.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
POST /api/local-model/install, cancel, resume and runtime, and DELETE
/api/local-model. The GET gains an `install` object the window polls. All of
it is experimental and admin only in multi-user mode.
One job of five steps (runtime, model, verify, agent, connect) driven by a
pure reducer over explicit state, step and transition tables. One job at a
time (409). The consent is bound to the size shown: `acknowledgedBytes` must
be the server's own total (409 size-changed). A job a server restart
interrupted comes back paused and never continues without a click. Every
failure is a code with one next step.
The model is judged against the budget the runtime measured before it is
downloaded. A missing coding agent is installed with its stock command
through the shared runner, audited. Connect writes one endpoint record under
the reserved id through mutateCustomModelHosts() and switches
customModelEndpointsEnabled on through mutateSettingsFile(), remembering the
flag's value. The model's digest is stored and read again at boot and before
every session start on that endpoint; a changed model refuses the launch.
Remove answers 409 with the session names while a live session uses the
endpoint, undoes the two writes first, restores the flag only when this
feature switched it on and no other endpoint is saved, and deletes only
inside local-model/. The coding agent stays.
The server starts the runtime at boot when a model is installed and stops
it at shutdown beside stopDeepSeekWeb().
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The four endpoint routes and the 5-minute discovery sweep each read
custom-model-hosts.json, changed their copy and wrote it back through one
shared temp file name. Two of them at once could tear the file or drop each
other's record: ten creates sent together did not all land.
Every write now goes through mutateCustomModelHosts() (one promise chain, a
random temp name, mode 0600), with each check inside the same step as its
write. A file that does not parse is refused with 409 and never replaced.
A delete of an unknown id no longer rewrites the file. A guard test fails
any other writer in src/.
The id `local-ollama` is reserved for the model Codeman runs on this Mac: a
new endpoint may not take it, and while the record is the managed one it is
not edited, deleted or re-discovered from these routes.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The runner that executes a CLI's stock install command (the environment
without CODEMAN_*, the npm prefix redirect, the bounded process group, the
one install per CLI at a time) moves from cli-registry-routes.ts to
src/cli-install-runner.ts, so the local model install can use the same one.
The route re-exports installEnv and npmGlobalPrefixWritable and behaves as
before; its tests are unchanged.
New in the shared module: an abort signal that kills the process group,
stockInstallCommand() (the compiled stock command, never the clis.json one)
and tryBeginCliInstall(), the one install slot per CLI for both callers.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
runtime.ts: the runtime is pinned in a compiled constant (version, URL, size,
sha256). The archive is streamed to a .partial file while it is hashed,
continued with a Range request, retried at most 3 times, and the hash is
compared before anything is unpacked; a mismatch deletes the file. It is
unpacked with /usr/bin/tar into a temporary folder and renamed.
The supervisor runs the runtime as a non-detached child with an environment
built from an allowlist by the pure runtimeEnv(), on a port chosen once by
binding. It restarts the child at most 3 times, stops it with SIGTERM, adopts
or replaces its own orphan on the port and answers port-in-use for any other
listener. parseInferenceComputeTotal() reads the device total the runtime
logs when it starts. Under vitest nothing spawns or fetches.
ollama-client.ts: the pull is parsed incrementally by a pure parser (a 23 GB
pull is about 1.4 million lines), with up to 5 retries on an error line or a
cut stream. verifyModel() loads the model with one tiny chat, checks what the
runtime reports (all on the GPU, the chosen context) and makes one tool
probe; each failure is a code.
developer-tools.ts: only /usr/bin/xcode-select -p, bounded, darwin only.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Codeman maintainer