verify
Evidence
verify_chain, security_scan, test_gap. No "trust me".
osien is a portable toolchain that installs itself into any AI. Forty-five tools for memory, verification, parallel work and context discipline — wired in one prompt. Small model or big, known or unknown: it works.
Your AI is only as good as the tools it can reach. Osien-Stack gives every AI the same unfair advantage, without you lifting a finger.
architecture, decisions, failures, state, performance. ctx_recall loads only what the role needs, so your AI never starts cold.
mtime-keyed file cache. 50 reads in 24ms. Tools return before you blink.
read_slice, smart_grep and ctx_recall feed the model only what matters. Stop paying for whole files.
verify_chain, security_scan, test_gap. No "trust me".
Mayor, Hammer, Detective debate in parallel.
opencode, Claude Code, Cursor, Cline, Windsurf.
Load project memory by role/part/grep.
Same AI. Same model. Different toolchain.
45 forged tools, short names, sharp descriptions. The model spends tokens on work, not on choosing. Browse all 45 →
soul_read, soul_write, fast_note, task_state, state_reset.
impact_analyze. Locations and lines, never whole files.
security_scan, test_gap, repro_capsule, context_budget, consensus, auto_heal.
repo_health, project_map, loop_status, checkpoint_create.
rollback, test_synthesizer, dead_code, dep_graph, changelog, token_budget and more.
city_status, city_post, city_read, city_broadcast, city_clear.
Run multiple read / grep / ctx ops in one call, in parallel. The powerhouse.
Token cost estimates and self-review loops before you ship.
A manifest goes in, finished work comes out. The batch runner handles the boring parts — concurrency, retries, duplicates, resume — so you just list the work.
A real run at full scale. 118 passed, 3 intentional failures, every one reported.
Concurrency is a ceiling, not a suggestion. 100+ tasks never become 100+ live processes.
Prove the whole plan at full scale before spending a single token.
It works through your task list in parallel, skips anything already done, retries whatever breaks, remembers exactly where it stopped, and hands you a pass/fail report at the end. Point it at 100+ jobs and it grinds through them while you get on with something else.
Task text is handed over as a single argument with shell=False, so metacharacters stay text and can never become commands. The AI binary is required through --bin and is never auto-guessed — a wrong guess silently runs the wrong program.
Click tools to chain them. Your AI does this automatically, but here you can feel it.
One MCP server over stdio. Any AI talks to it like a native tool. Your AI keeps its brain; Osien-Stack gives it hands.
One prompt: "install osien-stack". Clones, npm installs, wires the MCP server, copies the skills. Done.
45 tools register over stdio. Memory, verify, search, state, city mode, all callable by name.
SITE LAW, TOOL-FIRST and PARALLELISM are auto-loaded. It stops hand-rolling and starts forging.
93% fewer tokens, 0.49ms cached reads, parallel batch ops. Same model, ten times the throughput.
Your AI types the install command. You watch. That is the whole setup.
The questions every AI asks before installing.
Paste this into any AI. It does the rest. Any AI.