TypeChat

TypeChat

English · 简体中文

TypeChat is a library that makes it easy to build natural language interfaces using types.

Building natural language interfaces has traditionally been difficult. These apps often relied on complex decision trees to determine intent and collect the required inputs to take action. Large language models (LLMs) have made this easier by enabling us to take natural language input from a user and match to intent. This has introduced its own challenges including the need to constrain the model’s reply for safety, structure responses from the model for further processing, and ensuring that the reply from the model is valid. Prompt engineering aims to solve these problems, but comes with a steep learning curve and increased fragility as the prompt increases in size.

TypeChat replaces prompt engineering with schema engineering.

Simply define types that represent the intents supported in your natural language application. That could be as simple as an interface for categorizing sentiment or more complex examples like types for a shopping cart or music application. For example, to add additional intents to a schema, a developer can add additional types into a discriminated union. To make schemas hierarchical, a developer can use a “meta-schema” to choose one or more sub-schemas based on user input.

After defining your types, TypeChat takes care of the rest by:

1. Constructing a prompt to the LLM using types.
2. Validating the LLM response conforms to the schema. If the validation fails, repair the non-conforming output through further language model interaction.
3. Summarizing succinctly (without use of a LLM) the instance and confirm that it aligns with user intent.

Types are all you need!

Getting Started

The TypeScript/JavaScript package requires Node.js 24.22.1 or newer. Development uses Node.js 24.22.1, pinned in .nvmrc, which includes the October 21, 2026 security fixes.

Install TypeChat for TypeScript/JavaScript:

npm install typechat

You can also work with TypeChat from source for:

• Python
• TypeScript
• C#/.NET

To see TypeChat in action, we recommend exploring the TypeChat example projects. You can try them on your local machine or in a GitHub Codespace.

To learn more about TypeChat, visit the documentation which includes more information on TypeChat and how to get started.

Project direction

GOALS.md defines what belongs in TypeChat, what does not, and what is still under consideration. It is the primary repository policy for product scope. Changes require explicit approval from the human project owner, @robgruen; community interest and AI recommendations do not change it. Read it before proposing a new direction. Ordinary in-scope maintenance follows the normal contribution process.

Repository automation experiment

We are experimenting with a gradual transition to a semi-automated repository. The initial agent-assisted workflow keeps humans in the loop, while our longer-term goal is to help community proposals progress from discussion to accepted, reviewed changes with less manual coordination. Read about the automation experiment and how to participate.

Status as of 2026-11-03 (UTC): The latest backlog assessment completed successfully after the control hardening merged through #491. It inventories 20 issues and 13 open PRs. The earlier saved draft inventories 21 open issues and 25 open pull requests at the assessment time, with plain-language next steps and project-direction classifications.

The latest run used 23 model calls and 148.42 of the main assessment’s 400 AI credits. Detection used a separate 7.89 credits. Its 216 tool calls stayed within the allowed set; native diagnostics recorded one failure and no repeated requests. Verification remains partial: tests, CI definitions, and job logs were not checked. The earlier CI access restriction was not retested. Reports are not acceptance of work, GitHub Copilot code reviews, or merge approval.

The owner-approved discussion pilot has verified Issue draft generation, conversational /typechat reply delivery, and a duplicate-free reply retry after #498 merged. Manual response publication, Discussion replies, and Discussion acceptance still need live verification. /typechat draft remains preview-only, with exact-text response approval through /typechat publish. Owner-only /typechat accept and its no-op retry have also been verified on an Issue after #503 merged. Completion-only operational receipts were merged through #506; selected live checks verified Issue/Discussion input-needed draft receipts, invalid-command reporting, and controller-only no-op retries. Cancellation, uncertain-write recovery, acceptance-ready receipt wording, and direct-reply reuse remain unverified live. Receipts do not publish draft content or accept work. Generation keeps its 50-credit main budget and separately budgeted detection. Acceptance does not launch a coding agent or activate implementation spending.

The public-proposal extension is prepared for owner review and controlled live verification, not claimed live-verified. Its default flow is reply for conversation, or propose -> discuss/revise -> accept -> build -> PR -> review -> merge for development on an Issue. Proposals are posted directly for community discussion; the owner selects a visible version such as P2 for acceptance and a separately authorized build. IDs and hashes stay internal. The optional draft/publish review mode remains supported. See the owner cheat sheet.

The owner-triggered implementation pilot implements the workflow accepted in #504 as a separate, disabled-by-default extension. Review, activation, and a fresh owner build command remain distinct. The separately authorized #75 run created the App-authored #513, followed by normal CI, Copilot review, human fixes, and owner approval and merge. Original automated evidence remains bound to the original commit, not the human follow-ups. The activation flag is again unset; no coding run or merge happens automatically.

Milestone	Implemented	Current status
Community-facing automation plan	2026-10-26	Documented in AUTOMATION.md.
Manual backlog assessment, Copilot guidance, and workflow checks	2026-10-26	Merged via #482 on 2026-10-27 (UTC); manual assessment available on main.
Assessment model and reasoning effort	2026-10-26	GPT-6.1-sol and main-assessment max effort verified in the first live run on 2026-10-27 (UTC).
Reproducible workflow compilation	2026-10-27	Verified on GitHub: no generated-file drift and 17 passing tests.
First live backlog assessment	2026-10-27	Completed successfully; 24 issues and 26 PRs enumerated. PR investigation is partial; no dispositions are approved.
Clearer report language	2026-10-27	Writing guidance merged via #485. The successful October 29 report uses narrative entries and separates evidence from suggestions; human review remains required.
Assessment budget	2026-10-27	Main-assessment budget raised to 400 AI credits via #485, confirmed in the latest run. Detection remains separately budgeted.
Owner-controlled project direction	2026-10-27	Policy, assessment guidance, and CODEOWNERS merged via #485. The latest draft classifies every item against the pinned policy. Classifications are advisory; full GitHub enforcement still needs verification.
Call and execution-time limits	2026-10-29	1,000 calls, 60-minute agent execution, and a 100-minute enclosing job merged via #490 and used in the successful October 29 run. Credit budgets unchanged.
Successful policy-aware assessment	2026-10-29	Run details: 21 issues and 25 open PRs inventoried, with partial verification and human review still required.
Human follow-through	2026-10-29	Maintainers subsequently merged #433 and #437. These were human decisions, not automated acceptance.
Selected report archive	2026-10-30	The October 29 draft is preserved unchanged with provenance and review limits, for normal PR review. No automatic archiving is enabled.
Named project owner	2026-10-30	Policy and CODEOWNERS name @robgruen, merged through #491. Existing branch-review protections are unchanged.
Tool boundaries and run diagnostics	2026-10-30	Restricted model tools and native-only metrics merged through #491 and verified in the latest live run. CI access restrictions are not bypassed.
Owner-approved budget overrides	Not yet	Future work; no override input or approval mechanism is enabled.
Automatic Copilot PR reviews	2026-10-30	Active review rule for ready PRs targeting main and subsequent pushes; drafts are skipped. Configuration and the review on #491 were verified.
Stricter human-review enforcement	Not enabled	Existing branch-review protections are retained at the owner’s request. Copilot reviews do not replace human approval.
Owner-approved issue and Discussion drafts	2026-10-30	Merged through #495; run 41147057482 verified the Issue draft-only path. Publication and Discussion-path verification remain separate.
Owner-commanded direct replies	2026-10-31	Merged through #498. Attempt 1 posted one Issue reply; attempt 2 skipped inference and posting, preserving the original comment. Other paths remain unverified live.
Exact owner work acceptance	2026-10-31	Merged through #503. Attempt 1 recorded the approved Issue scope; attempt 2 preserved one unchanged record. Those acceptance commands did not start coding or activate spending; Discussion acceptance remains unverified live.
Completion receipts	2026-11-02	Merged through #506. Selected Issue/Discussion outcomes and controller retries are live-verified; other outcomes remain unverified live.
Acceptance and build acknowledgments	2026-11-02	Merged through #510. Issue #75 received a thumbs-up confirming exact acceptance and a rocket acknowledging build startup; Discussion acknowledgment remains unverified live.
Owner-triggered Issue implementation	2026-11-02	Merged through #507, default-off. The #75 run published #513 on 2026-11-03 (UTC); normal CI and Copilot reviews ran, human fixes were verified, and the owner approved and merged it. Original automated evidence and later human verification remain distinct; activation is again unset.
Selective automatic adoption of community proposals	Not yet	Future work, subject to public criteria and maintainer approval.

Implementation dates do not mean a feature is enabled on GitHub. Existing maintenance and deployment workflows are unchanged. See the detailed status and workflow model configuration.

The saved report is a historical snapshot, not today’s backlog. Later merges and decisions do not rewrite it or turn its recommendations into project policy.

Contributing

Start with the project goals and non-goals. Proposals that would change these boundaries need an owner-approved policy change before their implementation can be accepted.

Repository automation experiment: We are experimenting with a gradual transition to a semi-automated repository. The initial agent-assisted workflow keeps humans in the loop, while our longer-term goal is to help community proposals progress from discussion to accepted, reviewed changes with less manual coordination. Read about the automation experiment and how to participate.

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [email protected] with any additional questions or comments.

Website changes

The website source is in site. From the repository root:

npm --prefix site ci
npm --prefix site run build
npm --prefix site run serve

For UI changes, check keyboard navigation, including visible focus in Windows high-contrast themes. Focus indicators should have at least 3:1 contrast against the adjacent background. Check reflow at a 320 x 256 CSS-pixel viewport, equivalent to 400% zoom at 1280 x 1024. Also apply text-spacing overrides: line height of 1.5, letter spacing of 0.12em, word spacing of 0.16em, and paragraph spacing of 2em. The install command must remain readable and its copy button visible and keyboard-operable. With JavaScript disabled, the command should remain readable and the inactive copy button should be hidden.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft’s Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.