PrAImate 1.0.1

PrAImate 1.0.1

Fixes

• Studio Knowledge Tree: Fixed rendering of the root knowledge/ folder node and its children.
• State Sync: Fixed a stale UI state where the knowledge base would appear empty/missing immediately after enabling it or creating the first file.
• Build Pipeline: Assets now correctly include the recompiled praimate-gui binaries.

SHA256

c227b9c78120cdfbaa800087e2a42ed0435bffe594df739107635d621564f571  praimate-linux-amd64.tar.gz
142919f6936f3932d3becd59c331fd23a52bc13ea56afd49d66d9d4d7ae2f71c  praimate-linux-arm64.tar.gz
f4fcb5ef324e43eb11060c1949e42fb3c2ea02982a87cc35158c4383b4b97042  praimate-darwin-amd64.tar.gz
d10084803c2969000045dfd6c9d111bdfedabcd1a80b740385ccfa838a15ca42  praimate-darwin-arm64.tar.gz
4c85522201e5496cec0829b0fedb92b74288f44983c91e5cd88d63ddfb40abd5  praimate-windows-amd64.zip

PrAImate 1.0.0

PrAImate 1.0.0

One harness, every agent. PrAImate wraps the coding CLIs you already use — Claude Code, Codex CLI, OpenCode, Gemini CLI, DeepSeek-TUI, plus the bundled PrAImate Code — behind one launcher (TUI praimate, desktop GUI praimate --gui, and the wpc toolchain) with shared agents, memory, MCP, and automation.

Highlights

• Portable YAML agents + a curated starter set imported at first run.
• Agent authoring studio — an IDE for agents: a file tree of the agent folder, tabbed editor, knowledge base (Raw / RAG with graphify, live indexing progress), and an AI assistant whose CLI/model are switchable inline.
• Clean chats, code sessions & document Studio on any CLI/model — listed and reopenable from the Chats tab.
• Local LLM, end to end — inject a self-hosted endpoint at session creation, and route opencode/praimate-code, codex and deepseek to it from the Local LLM tab.
• Build bundled tools from source in Settings; the OpenCode fork + graphify sources are vendored in-repo.
• Cross-chat memory, MCP catalogue, folder watchers & cron, privacy redaction, optional git backup.
• Monke branding throughout (app icon, mascot, desktop shortcuts).

Install

# Linux / macOS
curl -fsSL https://raw.githubusercontent.com/sPROFFEs/PrAImate/main/scripts/install.sh | bash

# Windows
iwr -useb https://raw.githubusercontent.com/sPROFFEs/PrAImate/main/scripts/install.ps1 | iex

SHA256

2df06c3e955230c87d87649402d5d57ceabfde6acf87a206044d55b8d57f3f63  praimate-darwin-amd64.tar.gz
8712314401e3272f5ca6309c8c6b462de2fe39b48d7f54c10df2a1ee9b72b326  praimate-darwin-arm64.tar.gz
d15ef4847141831f2d371f5bc69a46a19f2beeb0cc7e2e2ef02894ed35c67c73  praimate-linux-amd64.tar.gz
b548643b0e3e7bf4d767ba6b6459ff5f2ed99485e9d02a443e3ce90b4dc4c10e  praimate-linux-arm64.tar.gz
9340edff640ee537c75138ff4c0dc7da0ec9b81ed9663b27c41421f62dd54b4d  praimate-windows-amd64.zip
49cb6a5a3839e1fb4e55600166daba1d540c261461170dc9964cb2c36c4d00bd  praimate-code-linux-amd64

Clade 0.1.18

Clade 0.1.18

• Hardened OpenClaude Local LLM routing with OpenClaude's native ~/.openclaude/.openclaude-profile.json openai profile instead of splitting HOME.
• Added Local LLM token cap settings so supported CLIs can stay inside smaller remote model limits.
• Added gstack as a Clade-managed Tool for supported host skill/slash-command workflows.
• Added ScrapeGraphAI as a Clade-managed search/extraction helper exposed as scrapegraph-search.
• Added _common/gstack and _common/scrapegraph bundles, with code-review enabling _common/scrapegraph as the example tool bundle.

Verify

sha256sum -c SHA256SUMS

### SHA256

6178c035abf3462e4e1b6d6316ecb2ff8cf9b776bf436af49bf932fd05d07c8e  clade-darwin-amd64.tar.gz
2b19261e202f131b3ee57171ee9709ff666ea336f942d371bef7b9b7b2361d1f  clade-darwin-arm64.tar.gz
730be1e5fa1781ff8a3c57d08d469367968767b790ba0284b9135ca85d776246  clade-linux-amd64.tar.gz
1190fc8c8076e5a4608339dc6c6934da9fb972acb50df3db1f8ed0b513ab0b1f  clade-linux-arm64.tar.gz
fb1c2b8734c963fd31a339577ccfa02232f4d0889db089066ad264a103e0183c  clade-windows-amd64.zip

0.1.15

Tools tab — install graphify (and future managed tools) from the TUI

Ctrl-4 opens a new Tools tab next to Agents. Today's only registered tool is graphify; rows show availability + version + install hint, Enter or i runs the install with live uv output streamed
inline. After install, the bin dir at <config>/clade/tools/graphify/bin/ is automatically prepended to PATH at every Clade startup (via ImportClademToolsToPath), so wpc-staged template scripts can call
graphify by name. Nav hotkeys shifted to make room: Backup is Ctrl-5, Local LLM Ctrl-6, Help Ctrl-7.

clade -install-tool <name> — same install path, headless

For scripts and one-shots. Same dispatch the TUI uses, no terminal UI:

clade -install-tool graphify

Unknown names print Available: graphify (and any future tools added to installer.KnownTools()).

On-launch hint when a bundle tool is missing

PrepareSandbox now walks every imported tool / agent / knowledge / hook's ImportedFrom, maps the bundle basename to a KnownTools() entry, and probes PATH + the managed prefix. If the underlying binary
isn't reachable, the launching screen surfaces a one-line note with the exact clade -install-tool <name> command. No silent "graphify: command not found" inside the chat.

Docs

README.md gains a top-level "Imports, managed tools, hooks (the 0.1.14+ feature set)" section: the two install paths for graphify (TUI tab + CLI), the hooks.json source schema with the portable-event →
claude mapping table, and a 3-step end-to-end verify block. Documentation moves from docs/ reference into the README where users actually land.

Carry-forward

All 0.1.14 features are still in: imports: in workpath.json, the managed-prefix install for graphify via uv tool install with UV_INDEX_URL pinned, hooks.json with seven portable events compiling to
.claude/settings.json for claude (other targets get a "declared but not wired" note). 0.1.13's supply-chain stance (per-pnpm --registry=https://registry.npmjs.org/, OpenClaude phantom-dep managed-prefix
workaround) remains in place.

SHA256

ad331cbf8d7c3c0863a60a0ab3d35d8964fbc1b126fc7f566a6851252377c55d clade-darwin-amd64.tar.gz
e566f2e53413d7b25db6330460a21dcc6dd2ef96c39f01e8a8dababd1df5d05b clade-darwin-arm64.tar.gz
88ecf143aa617ccaf6e26dd583b1be0955fe7c7504590b7f371e2e57d796e3d1 clade-linux-amd64.tar.gz
2896f7ff7dad7effba3d69d75fce589955d75636be5a54745a6ee60765c27315 clade-linux-arm64.tar.gz
22185d39a588bd6b43c0d5dedfc087e408fe7355c8c058c7b4c8b78f696443b8 clade-windows-amd64.zip

Clade 0.1.17

Clade 0.1.17 — local LLM caps and OpenClaude routing

• Added Local LLM context/output token caps, defaulting to 4096/1024.
• OpenCode writes model limit.context and limit.output into opencode.json.
• Codex writes the supported model context window setting.
• OpenClaude now uses the per-chat Local LLM endpoint with /v1 OpenAI-compatible env vars.
• OpenClaude now receives Local LLM token caps through its CLAUDE_CODE_OPENAI_* env maps.

Verify

sha256sum -c SHA256SUMS

### SHA256

908d2cb1105b5ee07b2b9de42213aa213239073a4023023f4986a875b581fdf4  clade-darwin-amd64.tar.gz
21a0cfb596ed10b62215822113a142485e43b0787613acc7f071ee8003957258  clade-darwin-arm64.tar.gz
e7ad9b972cac25ad7efc2ff6a7be979c2bb12edc574557fdc85c34045b99864b  clade-linux-amd64.tar.gz
f855fdb5a113e311034e0ed24773da1eb4c697aedd5a961508d2b4a5711a7a23  clade-linux-arm64.tar.gz
8e4f59167340b41391841066d3c2e0aff56a77f842ac5ff02524187695040d24  clade-windows-amd64.zip

Clade 0.1.16

Clade 0.1.16

• Adds the Tools tab and install-tool flow for Clade-managed tools.
• Adds chat-settings control for imported bundles.
• Improves probe/install progress rendering for slow agent CLIs.
• Fixes OpenCode false broken install status when opencode --version is slow to start.

SHA256

bcf4341aa695ae8559bfb754621b823fdc85eda9eeeae6efeef17c80ca73652d  clade-darwin-amd64.tar.gz
482880505675c04aa1d3777354088e6f8c36c2c2ae0437063a015225f8f1bdc1  clade-darwin-arm64.tar.gz
8d7705100f07158b2d06a8b2a32325361cf66fab53d3f94f407218c58caffff0  clade-linux-amd64.tar.gz
9a8103a492fa7c0547c69e6f58160b12432299d7be5e5dd618746b2f42a60f97  clade-linux-arm64.tar.gz
6c305b717d69f050782b80da4ead3295e82e70b375fe09e164398162012bb3e3  clade-windows-amd64.zip

0.1.14

wpc imports — shared capability bundles across templates

Templates can now declare "imports": ["_common/<bundle>"] in their workpath.json to pull in shared knowledge / tools / agents /
hooks / playbook+rules fragments from a sibling bundle. Collisions resolve in the consumer's favor; missing imports fail loud at compile
time. Nested templates (templates/<name>/workpath/) use "../_common/<bundle>" to escape the extra level. See docs/SCHEMA.md.

Clade-managed tools — first-class CLIs the agent calls

internal/installer grows a ToolID concept parallel to AgentID, with a managed install prefix under <config>/clade/tools/<name>/.
First registered tool: graphify, installed via uv tool install with the PyPI index pinned. Clade auto-adds each tool's bin dir to
PATH at startup so wpc-staged template scripts can call the tool by name. uv itself isn't auto-installed — Clade surfaces a clear
install command if it's missing (curl -LsSf https://astral.sh/uv/install.sh | sh etc).

Cross-harness hooks — write once, compile to each agent's native format

New hooks.json schema with seven portable events: pre_tool, post_tool, user_input, session_start, session_stop,
subagent_stop, notification. The claude target emits .claude/settings.json with the full Claude Code hook shape (per-event
groups, optional matcher, {type: command} entries). Targets without a stable hook system (codex, opencode, gemini, mika) append a
"hooks declared, NOT wired for X" note to their compiled instructions instead of silently dropping the hooks. Imports carry hooks with
the same collision policy as everything else (keyed on event+matcher).

Supply-chain stance

Carried forward from 0.1.13 and extended:

• Every pnpm add Clade emits includes --registry=https://registry.npmjs.org/.
• Every uv tool install Clade emits includes UV_INDEX_URL=https://pypi.org/simple/.
• The OpenClaude managed-prefix workaround (phantom @aws-sdk/client-bedrock-runtime dep) remains in place.
• The new tool installer uses the same managed-prefix isolation model.

Docs

• README.md — workpath section now mentions imports + hooks; DeepSeek's row fixed to codex; OpenClaude row added.
• docs/SCHEMA.md — new imports and hooks.json reference sections.

SHA256

0c8cd0f5c66721739086dd1ee9afa8ce771d5594653c8f0edd89da53d4abbeb5 clade-darwin-amd64.tar.gz
b7041afedba6545242e330ead851b30bf667e0a66451e1de290f6d001262803d clade-darwin-arm64.tar.gz
f9374ccc70d2c008785d3f0836a30f3b4430ed6569680ab0124e31c9123ec3dc clade-linux-amd64.tar.gz
bfd3c896c05514f4a02b36a45f248960ff18f57cb5d061cc26686749c5c7c00a clade-linux-arm64.tar.gz
31b9ae00a849eecab887febefffa5b06e0758fd9bbb1365fec826c38933c62cf clade-windows-amd64.zip

Clade 0.1.13

OpenClaude support

New sibling agent to Claude Code. OpenClaude is a claude-code fork that routes through any OpenAI-compatible endpoint (Ollama, GPUStack, GitHub Models, OAuth'd Codex, vLLM, LiteLLM, …) via CLAUDE_CODE_USE_OPENAI=1.

• Session store, --continue/--resume/--session-id/positional-prompt flags, and CLAUDE.md auto-discovery all inherited from upstream.
• Same wpc target as claude — CLAUDE.md + .claude/skills/<name>/SKILL.md scaffolding works unchanged.
• Local-endpoint wizard adds an openclaude checkbox; Plan() injects CLAUDE_CODE_USE_OPENAI=1 + OPENAI_BASE_URL/MODEL/API_KEY + --model on next launch.

OpenClaude install — phantom-dependency workaround

OpenClaude (0.13.0 + 0.14.0) statically imports @aws-sdk/client-bedrock-runtime but doesn't declare it (only @anthropic-ai/bedrock-sdk, which pulls it in transitively). That resolves under npm's flat hoisting but crashes under
strict pnpm with ERR_MODULE_NOT_FOUND on launch.

• Clade installs openclaude into a managed prefix (<config>/clade/agents/openclaude/) via a project-local pnpm add --config.node-linker=hoisted — the only mechanism that flat-hoists the transitive aws-sdk where openclaude's import
  resolves (pnpm add -g ignores every node-linker knob).
• The global pnpm config is untouched: codex/opencode/gemini/deepseek keep their strict global layout.
• ERR_PNPM_IGNORED_BUILDS (pnpm 10+ blocking sharp/protobufjs postinstall) is treated as success — build scripts stay blocked; openclaude doesn't need them at launch.

Global default local-LLM endpoint

New top-level Local LLM tab (^5) saves a default self-hosted endpoint (URL + API key + wire API) once, so you don't retype it on every chat.

• Edit endpoint / key / wire API, Test connection (queries the model list), Save as default, Clear.
• New chats that opt into a local endpoint offer "Use saved endpoint / Enter a new one"; "use saved" jumps straight to the live model query.
• Connection-only: model and per-agent selection stay per-chat, so each chat can still diverge. Already-configured chats are untouched.

Context primer fix

The fresh-launch context primer now points the agent at the compiled root doc that actually sits in its cwd (CLAUDE.md / AGENTS.md / GEMINI.md, which embed the mission/playbook/rules) plus MEMORY.md — instead of the standalone
playbook.md/rules.md, which live in the workpath source dir and are never copied into the sandbox.

Supply-chain hardening

• Every pnpm add Clade emits now carries an explicit --registry=https://registry.npmjs.org/. Defends against poisoned ~/.npmrc / project .npmrc / npm_config_registry redirection.
• OpenClaude is pnpm-only: no npm fallback, no curl|bash. Locked in by regression tests.
• --ignore-scripts deliberately not added globally: would break codex/opencode postinstall (native-binary fetch).

SHA256

f772d50328dbb72a6a8ac9249852b67c1a21a00a095aba67fe7cc57b68f19ac1 clade-darwin-amd64.tar.gz
a1284beec1660a2713dfc63e797795b21f966ff41866aa1c44a6e5c9d4aaa48d clade-darwin-arm64.tar.gz
6c2045caf5daacc82e39c0da475a03ddf81354614ccac1557838d2e9a7a1d4d7 clade-linux-amd64.tar.gz
4fbf7be4e4ba84fcab8665a38fcfc1ec65e8c2d65c844533ccfa054387b3f44b clade-linux-arm64.tar.gz
f2b6859ad514c28277c78346b84b72b026c083fd6e13dfa27efb8787d8a8c2aa clade-windows-amd64.zip

0.1.12

clade 0.1.12 — workpath injection, context primer, list UI polish

The point of a workpath is the agent actually follows it. 0.1.12
closes the gap between "Clade compiled mission/playbook/rules into
the sandbox" and "the agent loads them on every turn."

Workpath injection on every launch

• Claude now gets a root-level CLAUDE.md containing mission /
  playbook / rules / required-reading directive. Claude reads it
  unconditionally at session start. .claude/skills/<name>/SKILL.md
  still ships for the skill-discovery use case.
• DeepSeek-TUI switched from the claude wpc target to the codex
  target, so it auto-loads AGENTS.md at session start (its primary
  convention — .claude/skills/ was the fallback discovery path).
• Codex / OpenCode / Gemini already auto-loaded their root
  convention files (AGENTS.md / GEMINI.md); no change needed.

Context primer (new, on by default)

On every fresh launch, the launcher passes a short positional prompt
telling the agent to read MEMORY.md, playbook.md, and rules.md
and reply with "Context loaded — ready for your first task." before
waiting on the user. Fires on Claude / Codex / Gemini (their CLIs
accept a positional prompt for new chats). Skipped on OpenCode /
DeepSeek-TUI (their interactive modes don't accept a first-prompt
arg; both rely on the AGENTS.md auto-load above).

Suppressed on session resume — the agent already has prior
conversation state, so an extra "read MEMORY.md" instruction would
just be noise.

Per-chat toggle in chat settings → Context primer. Default ON;
disable when your agent's auto-load is enough.

Token estimate at launch

The launching screen now shows a rough token estimate for the
workpath injection:

Tokens: ~480 tokens (root 320 · memory 110 · primer 50 tokens) +14 KB knowledge (on demand)

Bytes / 4 heuristic. Off by ±20% in practice but useful to spot
runaway memory or unintentionally huge CLAUDE.md content before
you pay for it.

Chat + template list TUI polish

Less cramped. Non-selected chats: a single compact line. Selected
chat: indented details block with template / description /
resume-state summary, separated by blank lines from neighbours. A
horizontal rule sits above the "+ new chat" extras row so the list
groups cleanly visually.

Verify

sha256sum -c SHA256SUMS

f21888559afc0a6fff99fe9d97c4e9b72b33aec2739cf24245158860edbebc85 clade-darwin-amd64.tar.gz
4d3f885eaccba35739750e0fd5d337dada80853ef5581c6fd4f83593b832a324 clade-darwin-arm64.tar.gz
8a9fa0d8a42e2c888bd95e9401367d49ca13c1087a90eb19b977153e503f4c8b clade-linux-amd64.tar.gz
d7acbec6e5a9eebbc2f1d7ce23231c1f04e6aec667b20b588365fbdb95c2dedf clade-linux-arm64.tar.gz
8b3c29d1ba80c63f1744f9b13f3c35b1162ebcbcb61cb668b57482f0a751c377 clade-windows-amd64.zip

Compatibility

No schema breaks. Existing 0.1.11 configs round-trip without
migration; the new disableContextPrimer field defaults to false
(primer enabled). DeepSeek-TUI users will find a freshly-compiled
AGENTS.md at their chat sandbox root on next launch; the previous
.claude/skills/<name>/SKILL.md is left in place (harmless).

0.1.11

Your workspaces root becomes a git repository. Templates and chats —
including their MEMORY.md, captured transcripts, summaries, and the
full per-chat session slices — sync across machines. No GitHub /
Gitea REST API; just git over HTTPS or SSH with the credentials you
already have.

First run gets a 3-way fork

After picking the workspaces root, choose one of:

1. Empty folder — just chats/ + templates/.
2. Empty folder + bundled samples — same plus the
   reversing / code-review / workpath-author starters.
3. Clone from a git remote — pulls your existing chats and
   templates from a remote you've already pushed to from another
   machine. The remote must be public OR your git credentials
   (HTTPS credential helper / SSH key) must already be configured.
   On failure, Clade falls back to option 1 with a clear
   breadcrumb so you can link the remote later from the Backup tab.

Backup tab (Ctrl-4)

Header shows the current status (initialised / has-remote / clean
/ ahead / behind), last-sync timestamp, and this machine's ID.

Rows:

• Remote URL — edit the configured remote.
• Test connection — runs git ls-remote against the URL with a
  12s timeout. Surfaces "repository not found / private",
  "authentication failed", "network unreachable" in user-friendly
  language.
• Sync now — commits local changes, fetches, decides:
• In sync → nothing to do.
• Local ahead only → push.
• Remote ahead only → fast-forward pull.
• Diverged → opens the resolution popup.
• Reset from remote — DELETES local changes and matches the
  remote. Two consecutive y/n confirmations.
• Force push — overwrites the remote with local. One y/n
  confirmation.
• Disconnect — clears the remote URL and disables auto-sync.
  Local files are untouched.
• Auto-sync (opt-in) — fires on every Clade startup and exit.
• Force always local changes (opt-in, only with auto-sync) —
  bypasses the divergence popup; local always wins. LOUD warning
  on activation about two-machine races.

Divergence popup

When both sides have unique commits, the popup shows the lists of
local and remote commits with timestamps and offers four
reconciliations:

• [m] Merge — keeps both sides via a merge commit.
• [r] Rebase — replays local on top of remote.
• [p] Force push — discards remote, keeps local.
• [R] Reset — discards local, keeps remote.

Merge / rebase conflicts surface a clear panel with the conflicting
files and instructions to resolve via the shell.

Auto-sync safeguards

• Startup hook runs before the TUI; exit hook runs after Ctrl-C.
• Both perform safe operations only (commit + ff-push / ff-pull).
• Divergence halts with a stderr pointer to the Backup tab.
• Force-always-local is guarded by a per-commit Machine-ID
  trailer + a 24h window: if another machine pushed recently,
  Clade refuses to clobber and tells you to resolve manually. Even
  with the dangerous opt-in on, you can't silently lose work
  someone else just pushed.

MEMORY.md custom merge

MEMORY.md in any chat or template gets a custom merge driver
(clade --merge-memory) that concatenates both sides under a
"## --- merged from another machine at ---" separator instead
of producing <<<<<<< conflict markers. Registered automatically
via .gitattributes + .git/config on Init/Clone.

What gets tracked / ignored

.gitignore is managed by Clade:

• Everything at the workspaces root is ignored.
• chats/ and templates/ are explicitly un-ignored, and everything
  inside them is tracked — sandbox, native session slices,
  captured transcripts, all of it. Trade repo size for full-fidelity
  sync across machines.

If you've manually edited .gitignore, Clade detects the missing
managed marker and leaves your file alone.

Verify

sha256sum -c SHA256SUMS

6f85c3f2c382a3c4b0f38587d097b2f6fdb1a854888a99b9828727306a1fae24 clade-darwin-amd64.tar.gz
808e84f9eff638e10d858a625c664bcb733a97a6e320542f7ecfbd4f99c97b18 clade-darwin-arm64.tar.gz
c1ed1d508533b44bf27ee0fe1687a1f9de25c5cb988370293f4b03eaab2c407a clade-linux-amd64.tar.gz
dc9c9e6164dfb9551a0c1bfb4cca3aec61cf3f898ddc54655b6b5ae162bbd9aa clade-linux-arm64.tar.gz
d15c1373ef690b1cda2a7e412194dfcbe4d1fab8ce4bf076abb4c7c6de05e951 clade-windows-amd64.zip

Compatibility

No breaking changes. Existing 0.1.10 configs round-trip without
migration (the new backup fields are all omitempty). The first-run
flow only triggers on a fresh install; existing users see no
behavioral change until they explicitly open the Backup tab.