v0.4.1

[0.4.1] 2026-11-09

Bug fixes

• Current OpenAI and Anthropic models and request parameters for the hosted providers (#9)

Documentation

• Remove authorship statements from docs and scripts
• Recorded demos of more commands, with their vhs tapes (#2)
• Recorded demos of the HTTP service and more commands (#3)
• Complete citation metadata and bump it on release (#5)
• Render every architecture diagram in the house style (#7)
• Diagrams on white in the shared theme, make diagram for every diagram, and a self-contained Definition of Done (#8)

Build and CI

• Build and attach the sdist and wheel to every release (#10)

Maintenance

• Latest toolchain and CI versions (#4)
• Release script keeps hard-coded version strings in step (#6)

v0.4.0

[0.4.0] 2026-11-02

Features

• Per-tenant conversation memory with cited recall (#1)

Documentation

• Recorded terminal demo of a real run, with its vhs tape

Maintenance

• Classify squash merges in the changelog by the commits they contain

v0.3.0

[0.3.0] 2026-11-02

Features

• client: TypeScript client with streaming, job polling and typed errors

Documentation

• TypeScript client in README and diagram

v0.2.0

[0.2.0] 2026-11-02

Features

• providers: OpenAI and Anthropic providers with classified errors and failover
• api: Stream agent steps over server-sent events
• mcp: Serve the platform to MCP clients over stdio, add chat command

Documentation

• Standard README structure and language framing (#2)
• Vertical layered architecture diagram
• Horizontal layered architecture diagram
• Compact horizontal architecture diagram
• Providers, streaming, MCP and chat in README, diagram and agent notes
• List the provider and MCP modules in AGENTS.md and llms.txt

Build and CI

• Run CodeQL on main and on a weekly schedule only

Maintenance

• Add release automation, AGENTS.md, llms.txt and contributor docs

v0.1.0

First release of an integrated AI platform composed from the from-scratch libraries.

Highlights

• Authenticated, multi-tenant HTTP API with hashed API keys and per-tenant rate limits
• ReAct agent routed through the LLM gateway, with tenant-isolated hybrid retrieval and HNSW search
• Durable jobs in SQLite with leases and fencing tokens; a worker killed mid-job is replaced and resumes without re-running saved steps (tested with a real SIGKILL)
• Tracing, Prometheus metrics, and an evaluation suite: 95 percent accuracy and 100 percent citation rate on 20 questions

Install: pip install the attached wheel.