Metadata-Version: 2.4
Name: py-kodo
Version: 0.1.5
Summary: Kōdo (コード) — an open-source agentic harness
License:                                  Apache License
                                   Version 2.0, January 2004
                                http://www.apache.org/licenses/
        
           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
           1. Definitions.
        
              "License" shall mean the terms and conditions for use, reproduction,
              and distribution as defined by Sections 1 through 9 of this document.
        
              "Licensor" shall mean the copyright owner or entity authorized by
              the copyright owner that is granting the License.
        
              "Legal Entity" shall mean the union of the acting entity and all
              other entities that control, are controlled by, or are under common
              control with that entity. For the purposes of this definition,
              "control" means (i) the power, direct or indirect, to cause the
              direction or management of such entity, whether by contract or
              otherwise, or (ii) ownership of fifty percent (50%) or more of the
              outstanding shares, or (iii) beneficial ownership of such entity.
        
              "You" (or "Your") shall mean an individual or Legal Entity
              exercising permissions granted by this License.
        
              "Source" form shall mean the preferred form for making modifications,
              including but not limited to software source code, documentation
              source, and configuration files.
        
              "Object" form shall mean any form resulting from mechanical
              transformation or translation of a Source form, including but
              not limited to compiled object code, generated documentation,
              and conversions to other media types.
        
              "Work" shall mean the work of authorship, whether in Source or
              Object form, made available under the License, as indicated by a
              copyright notice that is included in or attached to the work
              (an example is provided in the Appendix below).
        
              "Derivative Works" shall mean any work, whether in Source or Object
              form, that is based on (or derived from) the Work and for which the
              editorial revisions, annotations, elaborations, or other modifications
              represent, as a whole, an original work of authorship. For the purposes
              of this License, Derivative Works shall not include works that remain
              separable from, or merely link (or bind by name) to the interfaces of,
              the Work and Derivative Works thereof.
        
              "Contribution" shall mean any work of authorship, including
              the original version of the Work and any modifications or additions
              to that Work or Derivative Works thereof, that is intentionally
              submitted to Licensor for inclusion in the Work by the copyright owner
              or by an individual or Legal Entity authorized to submit on behalf of
              the copyright owner. For the purposes of this definition, "submitted"
              means any form of electronic, verbal, or written communication sent
              to the Licensor or its representatives, including but not limited to
              communication on electronic mailing lists, source code control systems,
              and issue tracking systems that are managed by, or on behalf of, the
              Licensor for the purpose of discussing and improving the Work, but
              excluding communication that is conspicuously marked or otherwise
              designated in writing by the copyright owner as "Not a Contribution."
        
              "Contributor" shall mean Licensor and any individual or Legal Entity
              on behalf of whom a Contribution has been received by Licensor and
              subsequently incorporated within the Work.
        
           2. Grant of Copyright License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              copyright license to reproduce, prepare Derivative Works of,
              publicly display, publicly perform, sublicense, and distribute the
              Work and such Derivative Works in Source or Object form.
        
           3. Grant of Patent License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              (except as stated in this section) patent license to make, have made,
              use, offer to sell, sell, import, and otherwise transfer the Work,
              where such license applies only to those patent claims licensable
              by such Contributor that are necessarily infringed by their
              Contribution(s) alone or by combination of their Contribution(s)
              with the Work to which such Contribution(s) was submitted. If You
              institute patent litigation against any entity (including a
              cross-claim or counterclaim in a lawsuit) alleging that the Work
              or a Contribution incorporated within the Work constitutes direct
              or contributory patent infringement, then any patent licenses
              granted to You under this License for that Work shall terminate
              as of the date such litigation is filed.
        
           4. Redistribution. You may reproduce and distribute copies of the
              Work or Derivative Works thereof in any medium, with or without
              modifications, and in Source or Object form, provided that You
              meet the following conditions:
        
              (a) You must give any other recipients of the Work or
                  Derivative Works a copy of this License; and
        
              (b) You must cause any modified files to carry prominent notices
                  stating that You changed the files; and
        
              (c) You must retain, in the Source form of any Derivative Works
                  that You distribute, all copyright, patent, trademark, and
                  attribution notices from the Source form of the Work,
                  excluding those notices that do not pertain to any part of
                  the Derivative Works; and
        
              (d) If the Work includes a "NOTICE" text file as part of its
                  distribution, then any Derivative Works that You distribute must
                  include a readable copy of the attribution notices contained
                  within such NOTICE file, excluding those notices that do not
                  pertain to any part of the Derivative Works, in at least one
                  of the following places: within a NOTICE text file distributed
                  as part of the Derivative Works; within the Source form or
                  documentation, if provided along with the Derivative Works; or,
                  within a display generated by the Derivative Works, if and
                  wherever such third-party notices normally appear. The contents
                  of the NOTICE file are for informational purposes only and
                  do not modify the License. You may add Your own attribution
                  notices within Derivative Works that You distribute, alongside
                  or as an addendum to the NOTICE text from the Work, provided
                  that such additional attribution notices cannot be construed
                  as modifying the License.
        
              You may add Your own copyright statement to Your modifications and
              may provide additional or different license terms and conditions
              for use, reproduction, or distribution of Your modifications, or
              for any such Derivative Works as a whole, provided Your use,
              reproduction, and distribution of the Work otherwise complies with
              the conditions stated in this License.
        
           5. Submission of Contributions. Unless You explicitly state otherwise,
              any Contribution intentionally submitted for inclusion in the Work
              by You to the Licensor shall be under the terms and conditions of
              this License, without any additional terms or conditions.
              Notwithstanding the above, nothing herein shall supersede or modify
              the terms of any separate license agreement you may have executed
              with Licensor regarding such Contributions.
        
           6. Trademarks. This License does not grant permission to use the trade
              names, trademarks, service marks, or product names of the Licensor,
              except as required for reasonable and customary use in describing the
              origin of the Work and reproducing the content of the NOTICE file.
        
           7. Disclaimer of Warranty. Unless required by applicable law or
              agreed to in writing, Licensor provides the Work (and each
              Contributor provides its Contributions) on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
              implied, including, without limitation, any warranties or conditions
              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
              PARTICULAR PURPOSE. You are solely responsible for determining the
              appropriateness of using or redistributing the Work and assume any
              risks associated with Your exercise of permissions under this License.
        
           8. Limitation of Liability. In no event and under no legal theory,
              whether in tort (including negligence), contract, or otherwise,
              unless required by applicable law (such as deliberate and grossly
              negligent acts) or agreed to in writing, shall any Contributor be
              liable to You for damages, including any direct, indirect, special,
              incidental, or consequential damages of any character arising as a
              result of this License or out of the use or inability to use the
              Work (including but not limited to damages for loss of goodwill,
              work stoppage, computer failure or malfunction, or any and all
              other commercial damages or losses), even if such Contributor
              has been advised of the possibility of such damages.
        
           9. Accepting Warranty or Additional Liability. While redistributing
              the Work or Derivative Works thereof, You may choose to offer,
              and charge a fee for, acceptance of support, warranty, indemnity,
              or other liability obligations and/or rights consistent with this
              License. However, in accepting such obligations, You may act only
              on Your own behalf and on Your sole responsibility, not on behalf
              of any other Contributor, and only if You agree to indemnify,
              defend, and hold each Contributor harmless for any liability
              incurred by, or claims asserted against, such Contributor by reason
              of your accepting any such warranty or additional liability.
        
           END OF TERMS AND CONDITIONS
        
           APPENDIX: How to apply the Apache License to your work.
        
              To apply the Apache License to your work, attach the following
              boilerplate notice, with the fields enclosed by brackets "[]"
              replaced with your own identifying information. (Don't include
              the brackets!)  The text should be enclosed in the appropriate
              comment syntax for the file format. We also recommend that a
              file or class name and description of purpose be included on the
              same "printed page" as the copyright notice for easier
              identification within third-party archives.
        
           Copyright 2026 Stanislav Morozov
        
           Licensed under the Apache License, Version 2.0 (the "License");
           you may not use this file except in compliance with the License.
           You may obtain a copy of the License at
        
               http://www.apache.org/licenses/LICENSE-2.0
        
           Unless required by applicable law or agreed to in writing, software
           distributed under the License is distributed on an "AS IS" BASIS,
           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
           See the License for the specific language governing permissions and
           limitations under the License.
License-File: LICENSE
Requires-Python: >=3.12
Requires-Dist: aiohttp>=3.14.0
Requires-Dist: anthropic>=0.106.0
Requires-Dist: curl-cffi>=0.15.0
Requires-Dist: huggingface-hub>=1.18.0
Requires-Dist: nvidia-ml-py>=12.0.0
Requires-Dist: openai>=2.41.0
Requires-Dist: playwright>=1.61.0
Requires-Dist: psutil>=7.0.0
Requires-Dist: selectolax>=0.4.10
Description-Content-Type: text/markdown

# Kōdo

**Kōdo** (コード) is a build system that converts natural language into working code through a multi-agent LLM workflow — designed from the ground up to run on your own hardware, model included.

## Concept

Most AI coding tools live or die by the quality of your prompt — if you know exactly what you want and how to ask for it, they shine; if you don't, you get something that looks plausible but misses the mark. Kōdo is built on the belief that the bar shouldn't be that high.

Rather than expecting users to front-load all the right details, Kōdo asks. A structured multi-agent workflow interviews you, probes your goals, and surfaces the decisions you didn't know you needed to make — turning a rough idea into a rich, precise specification before a single line of code is written. Those specs are first-class artefacts, versioned and owned by you, living alongside your source code.

That's the first core idea: the spec is the source of truth. Kōdo's job is to help you build a good one, then translate it into running code automatically. Want to change how something works? Update the spec — Kōdo handles the rest.

The second core idea: none of this should require a subscription, an API key, or sending your code to someone else's datacenter. Kōdo runs as a Visual Studio Code extension talking to a local Python server — and that server can drive an open-weight model on your own GPU just as readily as a hosted API. With a cloud model, nothing leaves your machine except the LLM API call. With a local model, **nothing leaves your machine at all.**

To keep its promises verifiable, Kōdo currently focuses on backend software — logic, APIs, data pipelines — where correctness can be checked by tests without simulating a user interface.

## Local LLMs are first-class

Kōdo treats a GGUF running under llama.cpp exactly like a hosted API — same agents, same tools, same approval gates — and does the unglamorous work needed to make that actually true:

- **A curated model catalogue.** Two dozen ready-to-install builds across Qwen 3.6 (27B dense and 35B-A3B MoE), Qwen3-Coder-Next 80B, GPT-OSS 120B and 20B, Qwen 3.5 9B, Gemma 4 26B, and Ornith 1.0 35B — each entry carrying its quant spec, on-disk size, and hand-written hardware guidance for both discrete-GPU PCs and Apple Silicon Macs.

- **Hardware-fit detection.** Kōdo reads your GPU VRAM and system RAM and checks every catalogue entry against them *before* you download — a red warning when a build won't run on your machine, a yellow one when it will slow down at large contexts. The sizing accounts for llama.cpp's ability to split a model between GPU and system RAM (per-layer offloading for dense models, expert offloading for MoE models), so a modest consumer GPU is judged by what it can actually run — not by VRAM alone.

- **A real download manager.** Pause and resume that survive restarts and crashes, split-shard GGUFs handled automatically, live progress in every open window, and background failures surfaced instead of silently swallowed. See [`doc/LOCAL_MODEL_MANAGER.md`](doc/LOCAL_MODEL_MANAGER.md).

- **Agentic reliability hardening.** A local model's tool calls are parsed out of its raw token stream, and that format can slip in ways a hosted API's never does. Kōdo launches `llama-server` with grammar-constrained tool parsing, salvages tool calls the model emitted as plain text (behind a user confirmation), and strips stray `<think>` tags from reasoning — the difference between a demo and a 200-tool-call session that survives. See [`doc/LOCAL_INFERENCE.md`](doc/LOCAL_INFERENCE.md).

- **Bring your own.** Any HuggingFace GGUF, any local model file, or the URL of a llama-server you already run elsewhere — plus a binary override to point Kōdo at your own `llama-server` build.

Prefer a hosted model? Anthropic's model family is supported as a first-class alternative, with per-agent effort tiers and prompt caching keeping multi-agent token costs down. Either way, the server, your files, your specs, checkpoints, and session history stay on your machine.

Anthropic is the first cloud provider, not the last: support for OpenAI, Google, Meta, Alibaba, DeepSeek, and Kimi is planned, along with OpenRouter as an aggregator. Further out is a hybrid mode that treats local and cloud as one pool — a small local model triages each step of the workflow, escalating to a frontier model only when the work genuinely benefits from one and keeping everything routine on your own GPU.

## Small models, dense contexts

Are local models good enough for serious, multi-step engineering? Kōdo's architecture is built to make the answer yes — the multi-agent workflow isn't just quality control, it's context engineering for models that run on hardware you already own.

A monolithic coding agent accumulates one enormous transcript and re-reads all of it on every turn. Long context is exactly where local models hurt most: quality degrades, and the KV cache eats the very memory the weights need. Kōdo never puts any model in that regime. Each sub-agent starts from a clean context containing only the distilled inputs for its one job — the spec section under review, the test plan to implement, an investigation summary — does that job, returns a structured, schema-validated result, and exits. Nothing accumulates. Every context stays small and dense, which is precisely where a 9–80B open-weight model does its best work. When a long-running session does grow, context compaction kicks in automatically, sized to the active model's real context window.

The second half of the answer is verification. Kōdo writes tests from requirements before any implementation exists, pairs every author agent with a critic, and defines "done" as "the tests pass." That replaces *right on the first try* — what you pay frontier-model prices for — with *verifiably right after iteration*. On your own GPU, iteration costs electricity, not tokens.

## Two ways to work

**The guided pipeline** takes a green-field idea to a tested system through staged specification and review — the full workflow below.

**The Problem Solver** is the everyday entrance: point it at any codebase (Kōdo-built or not) and ask for a change, a fix, or a written investigation. It orchestrates dedicated Investigator, Planner, and Developer sub-agents for substantial work, and just does small asks directly — no ceremony for a one-file change.

The two are complementary rather than parallel: guided mode is planned project work, the Problem Solver is for what needs handling right now. The roadmap brings them together — the Problem Solver is set to learn how to operate on guided projects, updating the spec together with the code so an urgent fix never leaves the two out of sync.

## The guided workflow

1. **Init** — `Kodo: Init Project` lays down `.kodo/` (with the `kodo.md` manifest), `specs/`, `src/`, and `test/`.
2. **Prompt** — describe your idea in the WebView. The Narrative Author drafts a top-level description, optionally after a preliminary investigation of your existing code and the web.
3. **Architecture** — the Architect carves the work into components and emits a dependency graph used later for integration-test scheduling.
4. **Per-component specs** — for each component, author/critic pairs iterate on Requirements, then Functional Design, then a Test Plan, with an approval gate between every stage.
5. **Tests first** — the Test Coder produces failing tests from the test plan; nothing is implemented yet.
6. **Implementation** — the Coder iterates until every test passes; the Code Critic gates the diff.
7. **End-to-end** — where the Architect deems it applicable, an end-to-end test plan and suite assemble the whole system behind declared seams and verify it as a black box.
8. **Final approval** — the workflow closes and you review the full checkpoint history.

At each gate you can **Agree** (proceed) or **provide feedback** (re-run only the responsible author/critic pair with your input). A global **STOP** is available at all times. **Autonomous mode** runs the workflow unattended — agents resolve uncertainty with documented assumptions instead of questions, while the security layer stays live.

## A Kōdo project's structure

A Kōdo project treats `.kd` files as its primary source. Generated code lives alongside.

```text
specs/    *.kd files — narrative, responsibilities, per-component specs; the source of truth
src/      generated source code — all components and modules, including entry points
test/     generated unit tests, integration tests, and the end-to-end test
.kodo/    Kōdo working state — kodo.md manifest, checkpoint mirror (git), settings, logs
```

The relationship mirrors a traditional compiled project: specs are to generated code as source is to binary. Humans own `specs/` and approve everything that lands in `src/` and `test/`. For the MVP, `.kd` is plain Markdown — extended-tag variants are post-MVP.

## Key features

**Multi-agent workflow** — two entry agents (the Kōdo guide and the Problem Solver) orchestrate a roster of twenty-plus specialised sub-agents: authors paired with critics that gate output until quality is acceptable (capped at five iterations before escalating), plus standalone investigators, planners, developers, and toolchain agents.

**TDD by construction** — tests are written from requirements before any implementation exists. The Coder's loop terminates when tests pass; if a requirement is not testable, the Test Designer pushes back during specification, before code is written.

**Behaviour testing, not implementation testing** — generated tests assert observable outcomes rather than call counts or internal mocks. LLMs tend toward brittle, implementation-coupled tests; Kōdo's Test Design Critic enforces the opposite by default.

**Approval gates with feedback loops** — every stage ends at a gate. Agree to proceed, or provide feedback that re-runs only the responsible agent pair — never the entire workflow. No work lands in `src/` or `test/` until you approve.

**Full control over changes via the git mirror** — every mutating step lands as a commit in a shadow git mirror inside `.kodo/checkpoints/`, without touching (or requiring) a repository of your own. Roll the whole workspace back to any point in history — and forward again, nothing is destructive — or undo and redo an individual change's files independently, and diff between any two states. Nothing an agent does is beyond your reach to inspect or reverse. See [`doc/CHECKPOINTS.md`](doc/CHECKPOINTS.md).

**Crash-safe sessions** — every session persists as it runs and resumes exactly where it stopped, even mid-turn, after a crash or a window reload. Multiple sessions run in parallel tabs, across VS Code windows, against one shared local server. See [`doc/SESSIONS.md`](doc/SESSIONS.md).

**Security layer** — every tool call passes through a per-call allow-or-ask judgement driven by the Tool Control posture (permissive / defensive / smart). Smart mode statically analyzes shell commands for targets outside the workspace and runs an LLM intent judge over high-impact calls; anything it cannot clear raises a permission prompt. See [`doc/SECURITY.md`](doc/SECURITY.md).

**Web-capable research** — an agent-driven web search that paces its own discovery/read/synthesis loop, with browser-backed and static page extraction. See [`doc/WEB_SEARCH.md`](doc/WEB_SEARCH.md).

**Quantified quality control** — Kōdo's agent prompts are not tuned by gut feel. An internal harness, `kodo.validator`, drives real sessions through the real server, protocol, tools, and gates — no VS Code, no human — and records a complete transcript for scoring, giving system prompt changes a measurable quality signal. This matters doubly for local models: open-weight families differ enough in behaviour that a prompt that suits one can fail another, so the harness is what makes per-model system prompt curation tractable. See [`doc/VALIDATOR.md`](doc/VALIDATOR.md).

**Visual Studio Code extension** — streamed agent output, file diffs in VS Code's native diff editor, approval and permission prompts, cumulative cost, checkpoint controls, and a Local Inference Settings panel for browsing, downloading, and managing local models — all without leaving the IDE.

**Narrow extension surfaces** — LLM providers are plugins (Anthropic and llama.cpp today); agents and language toolchains are markdown-defined sub-agents, so adding a role or an ecosystem is a prompt plus a schema, not an engine change. See [`doc/ADDING_A_SUBAGENT.md`](doc/ADDING_A_SUBAGENT.md).

## Building the project

The project uses [hatch](https://hatch.pypa.io) for environment and build management.
The version scheme is `major.minor.patchb{build}` (e.g. `0.1.0b12`), where the build
number is stored in the `build_number` file and auto-incremented on release builds.

### Development cycle

```text
code change  →  build  →  test  →  commit  →  hatch run build  →  deploy wheel
```

| Command | What it does |
| --- | --- |
| `hatch run fmt` | Auto-format source with ruff. |
| `hatch run lint` | Lint source with ruff. |
| `hatch run typecheck` | Type-check with mypy. |
| `hatch run test` | Run the test suite with pytest. |
| `hatch run check` | Run fmt, lint, typecheck, and tests — no build. |
| `hatch build` | Quick wheel + sdist using the current version in `pyproject.toml`. Does **not** increment `build_number` or run checks. Use during development to verify the build. |
| `hatch run check-version` | Sync `__version__` in `__init__.py` from `pyproject.toml`. |
| `hatch run build` | Full release pipeline: stamp version, fmt, lint, typecheck, test, build, post-increment `build_number`. |

The `build_number` file contains the build number for the work currently in progress.
Commit your changes *before* running `hatch run build` — this way the committed source matches
the build number recorded in the repository. `hatch run build` is intended to be the final step
once code changes are done, tests are green, and everything is committed. It produces a numbered
wheel, then advances `build_number` so the repository is already pointing at the next iteration.

## Directory layout

```text
doc/        — Design docs (INTERNALS.md is the map)
scripts/    — Build tooling (pre/post build, version stamping)
src/        — Source code
test/       — Tests (unit tests, functional tests, etc)
```

## Status

Early-stage. The design is captured across [`doc/`](doc/) — start with [`doc/INTERNALS.md`](doc/INTERNALS.md).

Release is gated on demonstrated capability, not a feature list. The `kodo.validator` harness ([`doc/VALIDATOR.md`](doc/VALIDATOR.md)) runs real Kōdo sessions end-to-end — real server, real protocol, real tools and gates, no VS Code, no human — with local models under test. v1.0 ships when a battery of medium-to-high-complexity scenarios builds end-to-end inside that harness with all generated tests passing: an HTTP/HTTPS server in C++ or Rust, a transactional in-memory database, a distributed-consensus NoSQL key/value store, and more.

Current scope: backend software, green-field guided mode, Anthropic cloud or local llama.cpp models.
