openheim 0.3.0

A fast, multi-provider LLM agent runtime written in Rust
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107

# Changelog

## [0.3.0] - 2026-06-01

### Added

- **System identity (`system.md`)** — `~/.openheim/system.md` defines the agent's base identity and is injected into the system prompt on every session. `openheim init` creates a default file. The prompt is now structured: identity block first, then skills, separated by clear section headers.
- **`default_skills` in config** — new `default_skills` array in `config.toml` auto-loads a set of skills into every session without passing `--skills` each time. Per-session skills are merged on top; duplicates are removed with defaults appearing first.
- **`default_skills` in `OpenheimBuilder`** — `.default_skills(vec![...])` builder method brings the same control to programmatic embeddings.
- **Work-directory sandbox** — new `work_dir` field in `config.toml` restricts `read_file` and `write_file` to a directory tree. When unset, the directory from which openheim is invoked is used. Symlinks are followed and canonicalized so they cannot be used to escape the boundary.
- **Shell access control** — new `allow_shell` boolean in `config.toml` (default `true`). When `false`, the `execute_command` tool is removed from the tool list entirely — the LLM never sees it and cannot request it.
- **Builder methods for security controls** — `OpenheimClient::builder()` gains `.work_dir(path)` and `.allow_shell(bool)`. Both override the corresponding config-file values.
- **Cross-compilation config** — `Cross.toml` added for building Linux targets from macOS.

### Fixed

- **MCP subprocess stderr suppression** — stderr from spawned MCP server processes no longer leaks into the terminal.
- **`run` command exits cleanly** — the process now exits after a headless `openheim run` prompt completes instead of hanging.
- **`merge_skills` deduplication** — skills within the `default_skills` list itself are now deduplicated, not just across the default/session boundary.
- **Whitespace preserved in system identity** — leading and trailing whitespace in `system.md` content is preserved when building the system prompt.
- **Accurate `init` error message** — `openheim init` now correctly reports whether `system.md` was created when the config already exists.

### Breaking changes (library)

- `AppConfig` gained two new public fields: `work_dir: Option<PathBuf>` and `allow_shell: bool`. Code constructing `AppConfig` via struct literal (rather than TOML or the builder) must now supply these fields. Both have serde defaults so TOML loading is unaffected.
- `SystemToolExecutor::build` takes an additional `allow_shell: bool` argument.

## [0.2.1] - 2026-05-28

### Added

- github release workflow

## [0.2.0] - 2026-05-27

### Added

- **ratatui TUI** — complete rewrite of the terminal UI on ratatui; interactive picker for sessions, models, and commands replacing static text lists.
- **Token streaming & thinking blocks** — real token-level streaming and extended thinking display across all LLM providers (Anthropic, OpenAI, Gemini, OpenAI-compatible).
- **Themes** — built-in theme support; `:theme` command to switch at runtime.
- **`:models` command** — interactive popup picker to switch provider/model without leaving the session.
- **`session/set_model` in ACP** — clients can now switch the model mid-session via the Agent Client Protocol; thinking convention is advertised in session metadata.
- **`TerminalGuard` RAII** — terminal state is reliably restored on panic or forced exit.

### Fixed

- **MCP child process leak** — MCP server subprocesses are now killed when the client is dropped.
- **SSE buffer reallocation** — fixed a bug where the SSE read buffer could be reallocated incorrectly, silently dropping streamed tokens.
- **JSON serialization errors propagated** — LLM serialization failures now surface as errors instead of silently producing empty output.
- **Session restore** — restoring a session from history now correctly sets it as the active session and clears stale state.
- **Model switch shows full pair** — switching models now displays the full `(provider, model)` pair in the UI.
- **Provider config on session load** — ACP session load now resolves the full provider configuration instead of a partial view.
- **Model-whitelist enforced on provider resolution** — `resolve_with_provider` now checks the model whitelist before accepting a model.
- **Unconfigured provider warning on restore** — a clear warning is emitted when a restored session references a provider that is not in the current config.
- **Message word wrap** — long messages now wrap correctly in the TUI viewport.
- **Tool command error handling** — tool execution errors are captured and reported rather than silently dropped.

### Improved

- **Config, MCP, and skills views** — each view was refactored into a cleaner layout with consistent popup helpers.
- **Info panels moved to prompt area** — contextual info panels now render next to the input prompt instead of overlapping the message list.
- **TUI internals** — extracted `handle_scroll_key`, `highlight_row`, `centered_popup`, `push_screen`, and `Screen::is_overlay()` helpers; eliminated duplicated scroll/picker rendering code.

## [0.1.1] - 2026-05-21

### Fixed

- **Tool call history gaps** — ACP sessions were silently dropping tool calls from stored history; all tool calls are now captured correctly.
- **Accurate status on replay** — Replayed tool calls now emit `InProgress` before resolving, matching the behaviour of live sessions.
- **Failed tool calls now surface as `Failed`** — Previously, tool failures were stored as plain text and replayed as `Completed`. The `is_error` flag is now persisted in `Message` and propagated through `StreamEvent::ToolResult` so both live and replayed paths emit `ToolCallStatus::Failed`.
- **Tool error logging** — Improved logging for tool call errors.

### Improved

- **LLM accuracy on failures** — `is_error` is forwarded to Anthropic's `tool_result` block, giving the model accurate signal when a tool has failed.
- Added `CHANGELOG.md`.
- Updated documentation for `is_error` and tool call history replay semantics.
- README updates.

## [0.1.0] - 2026-05-15

First public release of openheim — a fast, multi-provider LLM agent runtime written in Rust.

### What's included

- **Multi-provider support** — OpenAI, Anthropic, Gemini, and any OpenAI-compatible endpoint
- **MCP integration** — connect external tools via Model Context Protocol (stdio and HTTP transports)
- **ACP server** — expose the agent over the Agent Client Protocol with WebSocket streaming
- **Tool execution** — built-in filesystem, shell, and extensible tool framework
- **Conversation history** — persistent sessions with RAG context and skill injection
- **Interactive REPL** — terminal UI for local development
- **Headless / programmatic mode** — embed openheim as a library in your own Rust application

### Install

```bash
cargo install openheim
```

Or add as a library:

```toml
[dependencies]
openheim = "0.1.0"
```

See the [README](https://github.com/weirdstuff-dev/openheim/blob/main/README.md) for configuration and usage.