Build with capsules

Build with structured context and graphs.

Research and architecture in development. This page explains our design direction. It is not a list of completed PRAXIS capabilities. See the documented product capabilities and open work.

Expected output. A visible failure.

For the pinned canonical fixture, the recorded structural result is valid=true, nine nodes and three edges. A missing referenced file or altered hashed passage must fail validation; stop and inspect the receipt rather than repairing meaning by inventing evidence. No hosted endpoint is established by installing the CLI. The browser replay uses the additive tacitus-web-context/1 view model; it does not migrate owner envelope schema 3 or capsule/v2.

Read the proposed adapter boundary and acceptance checks →

Download a reference capsule, inspect its graph and schema, then run the local compiler. DIALECTICA exposes Rust tooling, a CLI and Model Context Protocol (MCP) interfaces. Start from the pinned version and its prerequisites.

The open format is intended to make context inspectable across tools. Human review, source preservation, and versioned contracts guide the implementation. Hosted access and production integrations need a separate availability check.

Pinned local reference path

Source 3887db3 was inspected and its canonical fixture validated September 30, 2026. Install Git, rustup and the pinned Rust 1.95.0 toolchain; on Windows, use the MSVC build tools. The CLI can validate the bundled canonical fixture before using your documents. Fixture extraction is deterministic reference behavior; live provider integration and PRAXIS promotion need separate checks.

cargo run -p dialectica-cli -- validate fixtures/canonical-capsules/conflict-situation-capsule
cargo run -p dialectica-cli -- inspect fixtures/canonical-capsules/conflict-situation-capsule

Recorded result: valid=true, one sourced claim, nine graph nodes, three edges and a valid projection. Clone with core.autocrlf=false to preserve hashed fixture bytes on Windows. This is a format check, not a measured analytical-quality result. Preserve generated proposals for human review.

Option 1 — attach the MCP server to your agent

Clone the repository linked below and follow its setup instructions. Add the dialectica server to Claude Code or Codex, then begin with a small folder of public or non-sensitive documents.

# Claude Code (run from the cloned repository)
claude mcp add dialectica -- cargo run -p dialectica-mcp --

# Codex (config.toml)
[mcp_servers.dialectica]
command = "cargo"
args = ["run", "-p", "dialectica-mcp", "--"]
cwd = "/absolute/path/to/dialectica"
# Windows TOML: use forward slashes, e.g. C:/work/dialectica

Reference tool names to check against your repository version:

· dialectica_build_capsule — compile a capsule from a folder of your documents
· dialectica_inspect_capsule — summarize any bundle
· dialectica_validate_capsule — check it against the spec
· dialectica_ladybug_query — query the embedded graph with Cypher
· dialectica_export_praxis_pack — emit an agent-ready context pack
· dialectica_archive_capsule — write the portable .capsule archive

The technical idea

Give the next task a record
it can reason from.

Retrieval finds relevant material. Structure makes its meaning, relationships, and review state explicit. Explore how the parts are intended to work together.

Your evidence

Keep a claim close to its source.

Documents, passages, authors, and dates are the starting point. A statement should retain its origin and access boundary as it moves into structured knowledge.

Receives: Approved documents and a defined question
Produces: Source records with stable references
Quality question: Can a reviewer recover the passage and its context?

Illustrative record · Sources

source id: source-01
passage: The extension has been requested, not approved.
record type: statement
access: approved_for_this_task

Inspect the illustrative JSON

{
  "source_id": "source-01",
  "passage": "The extension has been requested, not approved.",
  "record_type": "statement",
  "access": "approved_for_this_task"
}

Teaching example. This is not a production schema or an API response.

Architecture under development. Integration and portability require end-to-end validation.

Current capabilities

Option 2 — the CLI

# clone and install the open-source engine
git -c core.autocrlf=false clone https://github.com/sargonxg/A3_DIALECTICAbyTACITUS_v3 dialectica
cd dialectica
git checkout 3887db306a815daaf4d30d40386efdefcf729f88
# Pinned rust-toolchain.toml: Rust 1.95.0; MSRV 1.81
cargo install --locked --path crates/dialectica-cli

# build a capsule from your own documents
dialectica build-docs --input ./my-documents --out ./my-capsule --type situation

# inspect, validate, archive
dialectica inspect ./my-capsule
dialectica validate ./my-capsule
dialectica archive ./my-capsule --out my-first.capsule.zip

Check the repository for the supported Rust toolchain and optional dependencies. These commands target a local workflow. Review provider settings before processing sensitive material; connecting an external model or service changes where data is sent.

The spec

Capsule spec v3

Nine layers, four types, one container format (mimetype-first zip). Canonical in the engine repository.

24 JSON Schemas

Exported by dialectica schema-export — manifests, ledgers, contracts, review decisions, registry listings.

Reference artifacts

A downloadable reference capsule plus one example file per capsule type.

All open-source repos Questions? One inbox, one founder