Skip to content

Codira Development Roadmap

Status

This is the versioned execution plan approved on 2026-08-02. It orders work by adoption value and dependency: agent access, evidence, mainstream coverage and freshness, productization, then differentiated and long-tail capabilities.

Local deterministic batch indexing remains canonical. Daemons, vector databases, shared services, and semantic retrieval are optional deployment modes; provenance and deterministic facts remain authoritative.

Completed Foundations

Completed before this roadmap: configuration and plugin configuration (#17,

27); documentation retrieval (#3); embedding calibration (#28); signal

aggregation (#32); plugin/analyzer contracts and concurrency (#4, #6, #55).

Deferred Design: Backend Write Concurrency

Tracked by: #56 Status: deferred; not scheduled and not a release gate.

Reopen only after either an approved genuinely multi-writer server backend, or a reproducible persistence bottleneck plus a staged prototype that improves end-to-end time by at least 20% while preserving determinism and reader isolation. A future design must preserve the codira.storage / backend-plugin boundary for lifecycle paths, metadata, run identity, and activation state.

Execution Sequence

  1. Agent-ready local MCP and automatic freshness.
  2. Reproducible comparative evidence.
  3. JavaScript, TypeScript/TSX, Go, and federation readiness.
  4. Productization and a gated shared-service decision.
  5. Rust, Java, PHP, architecture intelligence, and SARIF.
  6. Universal structural/text baseline, fragments, and build/infrastructure.
  7. Second-wave and long-tail language coverage.

Blocked work is scheduled after its blockers: #33 after #18; #1 after #67 and

35; #51 after #15, #20, #22, and #68.

v1.70.0 — Agent-ready local MCP

Window: 2026-08-03 to 2026-10-02

Deliver a usable local stdio MCP server, not only a design specification.

  • 58 MCP epic

  • 22 daemon mode for local watching, incremental refresh, and freshness

  • 61 roadmap/milestone reconciliation

  • 62 versioned MCP contract

  • 63 local stdio server and direct-core adapter

  • 64 trusted roots and output budgets

  • 65 client presets and integration suite

  • 66 compact agent-oriented repository map

Exit criteria: local read-only MCP has deterministic direct-core tools, authorized repository roots, bounded results, explicit freshness, and tested client setup. Multi-repository and shared service are explicitly out of scope.

v1.80.0 — Evidence and Comparative Benchmarks

Window: 2026-10-05 to 2026-11-20

  • 53 agent-efficiency benchmark suite

  • 59 embedding retrieval quality benchmark

  • 68 evidence gates for optional vector and shared-service deployment

Exit criteria: public and private reproducible fixtures, deterministic task oracles, task-success-first reporting, native and Codira baselines, and a documented gate for #20 and #51.

v1.90.0 — Mainstream Coverage and Federation Readiness

Window: 2026-11-23 to 2027-01-29

  • 36 JavaScript analyzer

  • 37 TypeScript and TSX analyzer

  • 38 Go analyzer

  • 15 multi-repository aggregation

  • 20 vector-backend gate decision

The #20 gate follows #59 and #68. Implement a new vector backend only when the benchmark demonstrates a material recall, latency, or memory limit in an important repository class; otherwise convert #20 to deferred with that gate.

v2.0.0 — Product Platform and Shared-Service Decision

Window: 2027-02-01 to 2027-04-30

  • 18 plugin extraction readiness

  • 33 final package split and publish rehearsal, after #18

  • 34 trusted publishing verification, after #33

  • 51 shared repository index service gate/implementation, including the

    deferred multi-repository and daemon-backed MCP continuation from #58

  • 69 productization baseline and support-tier contract

The #51 gate follows #15, #20, #22, and #68. It requires stable MCP, verified clients, benchmark coverage, and demonstrated multi-repository/local-index adoption; otherwise convert #51 to deferred with those reopening criteria.

v2.10.0 — Mainstream Expansion and Repository Intelligence

Window: 2027-05-03 to 2027-07-30

  • 40 Rust analyzer

  • 39 Java analyzer

  • 43 PHP analyzer

  • 70 semantic JSON document classification and manifest extraction

  • 54 complete repository architecture report

  • 52 SARIF output

66 is the early agent-facing slice of #54. #70 extends the existing JSON

analyzer, rather than adding a parser-level JSON-manifest format, so its composable document facets and bounded manifest facts are available to #54. This release completes dependency graphs and cycles, hotspots, layer rules, DOT/SVG rendering, and the human report. SARIF follows rather than precedes agent adoption and benchmark proof.

v2.20.0 — Universal Baseline

Window: 2027-08-02 to 2027-10-29

  • 5 Makefile analyzer

  • 12 Lua analyzer

  • 42 YAML/TOML/HCL analyzer

  • 67 structural and text fallback baseline

  • 35 fragment delegation

  • 14 TeX/LaTeX analyzer

  • 1 optional fallback analyzers, after #67 and #35

Support tiers are explicit: Tier A is language semantic; Tier B is structural; Tier C is text retrieval. Fragment recovery never replaces authoritative primary-parser facts.

v2.30.0 — Second-Wave Language Coverage

Window: 2027-11-01 to 2028-01-28

  • 44 Kotlin analyzer

  • 49 Ruby analyzer

  • 41 C# analyzer

v2.40.0 — Long-Tail Language Coverage

Window: 2028-01-31 to 2028-05-26

  • 45 Assembly analyzer

  • 46 Fortran analyzer

  • 48 Ada analyzer

  • 50 VHDL analyzer

  • 47 COBOL analyzer

Planning Rules

  • Releases use a roughly four-week cadence only when their scope permits it; the larger integration releases above deliberately have longer windows.
  • Every open issue is assigned to a release or a scheduled gate decision.
  • Only #56 is excluded from operational scheduling because it already has explicit deferred reopening gates.
  • New analyzer work must remain a separate plugin conforming to the existing contract, with deterministic fixtures and explicit supported capabilities.